Churn_Prediction_para_uma_empresa_de_Telecomunicações.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "Churn Prediction para uma empresa de Telecomunicações.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "include_colab_link": true
    },
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/JoaoGustavo29/Portfolio_DataScience/blob/master/Churn_Prediction_para_uma_empresa_de_Telecomunica%C3%A7%C3%B5es.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "sC2BFMOKN1M7"
      },
      "source": [
        "---\n",
        "<img alt=\"Colaboratory logo\" width=\"15%\" src=\"https://raw.githubusercontent.com/carlosfab/escola-data-science/master/img/novo_logo_bg_claro.png\">\n",
        "\n",
        "#### **Data Science na Prática 2.0**\n",
        "*by [sigmoidal.ai](https://sigmoidal.ai)*  \n",
        "\n",
        "---"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "IR56trFcPcC1"
      },
      "source": [
        "# Churn Prediction\n",
        "\n",
        "*Churn rate*, ou simplesmente *churn*, representa a taxa de evasão da sua base de clientes. Em serviços como Spotify ou Netflix, ela representaria a taxa de cancelamento de assinaturas.\n",
        "\n",
        "<p align=center>\n",
        "<img src=\"https://raw.githubusercontent.com/carlosfab/dsnp2/master/img/churnrate.jpg\" width=\"60%\"></p>\n",
        "\n",
        "Ela é de extrema importância para a administração e sua análise ao longo do tempo pode mostrar que existe algum problema que deve ser atacado.\n",
        "\n",
        "*Churn* também pode ser usado para identificar potenciais cancelamentos, com um tempo de antecedência, e promover ações direcionadas para tentar reter tais clientes. Essa métrica deve receber atenção pelo fato de que o Custo de Aquisição de Cliente (CAC) é normalmente mais alto que o custo para mantê-los. Ou seja, um alto valor para o *churn rate* é o que não desejamos."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "0HVmpIxQWT4Y"
      },
      "source": [
        "## Aquisição dos Dados\n",
        "\n",
        "Os dados utilizados neste projeto foram originalmente disponibilizados na [plataforma de ensino da IBM Developer](https://developer.ibm.com/technologies/data-science/patterns/predict-customer-churn-using-watson-studio-and-jupyter-notebooks/#), e tratam de um problema típico de uma companhia de telecomunicações. O *dataset* completo pode ser encontrado [neste link](https://raw.githubusercontent.com/carlosfab/dsnp2/master/datasets/WA_Fn-UseC_-Telco-Customer-Churn.csv).\n",
        "\n",
        "Apesar de não haver informações explícitas disponíves, os nomes das colunas permitem um entendimento a respeito do problema."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "_Zuw7p0Jcu_D"
      },
      "source": [
        "# instalar pacotes necessários\n",
        "! pip install -q scikit-plot"
      ],
      "execution_count": 1,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 386
        },
        "id": "oQ4bNy7udtEE",
        "outputId": "b10ba628-3364-44be-b7d9-b1df174bb2a5"
      },
      "source": [
        "# importar os pacotes necessários\n",
        "import pandas as pd\n",
        "import matplotlib.pyplot as plt\n",
        "import seaborn as sns\n",
        "import numpy as np\n",
        "import scikitplot as skplt\n",
        "from imblearn.under_sampling import RandomUnderSampler\n",
        "from imblearn.over_sampling import SMOTE, ADASYN\n",
        "from sklearn.model_selection import train_test_split, GridSearchCV\n",
        "from sklearn.model_selection import StratifiedKFold\n",
        "from sklearn.model_selection import cross_val_score\n",
        "from sklearn.pipeline import make_pipeline\n",
        "from sklearn.preprocessing import StandardScaler, RobustScaler\n",
        "from sklearn.preprocessing import QuantileTransformer, PowerTransformer\n",
        "from sklearn.preprocessing import LabelEncoder, OneHotEncoder\n",
        "from sklearn.metrics import confusion_matrix, classification_report\n",
        "from sklearn.metrics import roc_auc_score, recall_score\n",
        "from sklearn.ensemble import VotingClassifier, RandomForestClassifier\n",
        "from sklearn.linear_model import LogisticRegression, SGDClassifier\n",
        "from sklearn.discriminant_analysis import LinearDiscriminantAnalysis\n",
        "from sklearn.neighbors import KNeighborsClassifier\n",
        "from sklearn.svm import SVC\n",
        "from xgboost import XGBClassifier\n",
        "\n",
        "# importar Warnings filter\n",
        "from warnings import simplefilter\n",
        "\n",
        "# ignorar futuros warnings\n",
        "simplefilter(action='ignore', category=FutureWarning)\n",
        "\n",
        "# configurar notebook\n",
        "sns.set_style('dark')\n",
        "\n",
        "# importar os dados\n",
        "DATA_PATH = \"https://raw.githubusercontent.com/carlosfab/dsnp2/master/datasets/WA_Fn-UseC_-Telco-Customer-Churn.csv\"\n",
        "df = pd.read_csv(DATA_PATH)\n",
        "\n",
        "# ver as 5 primeiras entradas\n",
        "df.head()"
      ],
      "execution_count": 2,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "   customerID  gender  SeniorCitizen Partner Dependents  tenure PhoneService  \\\n",
              "0  7590-VHVEG  Female              0     Yes         No       1           No   \n",
              "1  5575-GNVDE    Male              0      No         No      34          Yes   \n",
              "2  3668-QPYBK    Male              0      No         No       2          Yes   \n",
              "3  7795-CFOCW    Male              0      No         No      45           No   \n",
              "4  9237-HQITU  Female              0      No         No       2          Yes   \n",
              "\n",
              "      MultipleLines InternetService OnlineSecurity  ... DeviceProtection  \\\n",
              "0  No phone service             DSL             No  ...               No   \n",
              "1                No             DSL            Yes  ...              Yes   \n",
              "2                No             DSL            Yes  ...               No   \n",
              "3  No phone service             DSL            Yes  ...              Yes   \n",
              "4                No     Fiber optic             No  ...               No   \n",
              "\n",
              "  TechSupport StreamingTV StreamingMovies        Contract PaperlessBilling  \\\n",
              "0          No          No              No  Month-to-month              Yes   \n",
              "1          No          No              No        One year               No   \n",
              "2          No          No              No  Month-to-month              Yes   \n",
              "3         Yes          No              No        One year               No   \n",
              "4          No          No              No  Month-to-month              Yes   \n",
              "\n",
              "               PaymentMethod MonthlyCharges  TotalCharges Churn  \n",
              "0           Electronic check          29.85         29.85    No  \n",
              "1               Mailed check          56.95        1889.5    No  \n",
              "2               Mailed check          53.85        108.15   Yes  \n",
              "3  Bank transfer (automatic)          42.30       1840.75    No  \n",
              "4           Electronic check          70.70        151.65   Yes  \n",
              "\n",
              "[5 rows x 21 columns]"
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-156f3473-3408-442f-b1b8-7131c691a669\">\n",
              "    <div class=\"colab-df-container\">\n",
              "      <div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>customerID</th>\n",
              "      <th>gender</th>\n",
              "      <th>SeniorCitizen</th>\n",
              "      <th>Partner</th>\n",
              "      <th>Dependents</th>\n",
              "      <th>tenure</th>\n",
              "      <th>PhoneService</th>\n",
              "      <th>MultipleLines</th>\n",
              "      <th>InternetService</th>\n",
              "      <th>OnlineSecurity</th>\n",
              "      <th>...</th>\n",
              "      <th>DeviceProtection</th>\n",
              "      <th>TechSupport</th>\n",
              "      <th>StreamingTV</th>\n",
              "      <th>StreamingMovies</th>\n",
              "      <th>Contract</th>\n",
              "      <th>PaperlessBilling</th>\n",
              "      <th>PaymentMethod</th>\n",
              "      <th>MonthlyCharges</th>\n",
              "      <th>TotalCharges</th>\n",
              "      <th>Churn</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>7590-VHVEG</td>\n",
              "      <td>Female</td>\n",
              "      <td>0</td>\n",
              "      <td>Yes</td>\n",
              "      <td>No</td>\n",
              "      <td>1</td>\n",
              "      <td>No</td>\n",
              "      <td>No phone service</td>\n",
              "      <td>DSL</td>\n",
              "      <td>No</td>\n",
              "      <td>...</td>\n",
              "      <td>No</td>\n",
              "      <td>No</td>\n",
              "      <td>No</td>\n",
              "      <td>No</td>\n",
              "      <td>Month-to-month</td>\n",
              "      <td>Yes</td>\n",
              "      <td>Electronic check</td>\n",
              "      <td>29.85</td>\n",
              "      <td>29.85</td>\n",
              "      <td>No</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>5575-GNVDE</td>\n",
              "      <td>Male</td>\n",
              "      <td>0</td>\n",
              "      <td>No</td>\n",
              "      <td>No</td>\n",
              "      <td>34</td>\n",
              "      <td>Yes</td>\n",
              "      <td>No</td>\n",
              "      <td>DSL</td>\n",
              "      <td>Yes</td>\n",
              "      <td>...</td>\n",
              "      <td>Yes</td>\n",
              "      <td>No</td>\n",
              "      <td>No</td>\n",
              "      <td>No</td>\n",
              "      <td>One year</td>\n",
              "      <td>No</td>\n",
              "      <td>Mailed check</td>\n",
              "      <td>56.95</td>\n",
              "      <td>1889.5</td>\n",
              "      <td>No</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>3668-QPYBK</td>\n",
              "      <td>Male</td>\n",
              "      <td>0</td>\n",
              "      <td>No</td>\n",
              "      <td>No</td>\n",
              "      <td>2</td>\n",
              "      <td>Yes</td>\n",
              "      <td>No</td>\n",
              "      <td>DSL</td>\n",
              "      <td>Yes</td>\n",
              "      <td>...</td>\n",
              "      <td>No</td>\n",
              "      <td>No</td>\n",
              "      <td>No</td>\n",
              "      <td>No</td>\n",
              "      <td>Month-to-month</td>\n",
              "      <td>Yes</td>\n",
              "      <td>Mailed check</td>\n",
              "      <td>53.85</td>\n",
              "      <td>108.15</td>\n",
              "      <td>Yes</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>7795-CFOCW</td>\n",
              "      <td>Male</td>\n",
              "      <td>0</td>\n",
              "      <td>No</td>\n",
              "      <td>No</td>\n",
              "      <td>45</td>\n",
              "      <td>No</td>\n",
              "      <td>No phone service</td>\n",
              "      <td>DSL</td>\n",
              "      <td>Yes</td>\n",
              "      <td>...</td>\n",
              "      <td>Yes</td>\n",
              "      <td>Yes</td>\n",
              "      <td>No</td>\n",
              "      <td>No</td>\n",
              "      <td>One year</td>\n",
              "      <td>No</td>\n",
              "      <td>Bank transfer (automatic)</td>\n",
              "      <td>42.30</td>\n",
              "      <td>1840.75</td>\n",
              "      <td>No</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>9237-HQITU</td>\n",
              "      <td>Female</td>\n",
              "      <td>0</td>\n",
              "      <td>No</td>\n",
              "      <td>No</td>\n",
              "      <td>2</td>\n",
              "      <td>Yes</td>\n",
              "      <td>No</td>\n",
              "      <td>Fiber optic</td>\n",
              "      <td>No</td>\n",
              "      <td>...</td>\n",
              "      <td>No</td>\n",
              "      <td>No</td>\n",
              "      <td>No</td>\n",
              "      <td>No</td>\n",
              "      <td>Month-to-month</td>\n",
              "      <td>Yes</td>\n",
              "      <td>Electronic check</td>\n",
              "      <td>70.70</td>\n",
              "      <td>151.65</td>\n",
              "      <td>Yes</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "<p>5 rows × 21 columns</p>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-156f3473-3408-442f-b1b8-7131c691a669')\"\n",
              "              title=\"Convert this dataframe to an interactive table.\"\n",
              "              style=\"display:none;\">\n",
              "        \n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "       width=\"24px\">\n",
              "    <path d=\"M0 0h24v24H0V0z\" fill=\"none\"/>\n",
              "    <path d=\"M18.56 5.44l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94zm-11 1L8.5 8.5l.94-2.06 2.06-.94-2.06-.94L8.5 2.5l-.94 2.06-2.06.94zm10 10l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94z\"/><path d=\"M17.41 7.96l-1.37-1.37c-.4-.4-.92-.59-1.43-.59-.52 0-1.04.2-1.43.59L10.3 9.45l-7.72 7.72c-.78.78-.78 2.05 0 2.83L4 21.41c.39.39.9.59 1.41.59.51 0 1.02-.2 1.41-.59l7.78-7.78 2.81-2.81c.8-.78.8-2.07 0-2.86zM5.41 20L4 18.59l7.72-7.72 1.47 1.35L5.41 20z\"/>\n",
              "  </svg>\n",
              "      </button>\n",
              "      \n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      flex-wrap:wrap;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "      <script>\n",
              "        const buttonEl =\n",
              "          document.querySelector('#df-156f3473-3408-442f-b1b8-7131c691a669 button.colab-df-convert');\n",
              "        buttonEl.style.display =\n",
              "          google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "        async function convertToInteractive(key) {\n",
              "          const element = document.querySelector('#df-156f3473-3408-442f-b1b8-7131c691a669');\n",
              "          const dataTable =\n",
              "            await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                     [key], {});\n",
              "          if (!dataTable) return;\n",
              "\n",
              "          const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "            '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "            + ' to learn more about interactive tables.';\n",
              "          element.innerHTML = '';\n",
              "          dataTable['output_type'] = 'display_data';\n",
              "          await google.colab.output.renderOutput(dataTable, element);\n",
              "          const docLink = document.createElement('div');\n",
              "          docLink.innerHTML = docLinkHtml;\n",
              "          element.appendChild(docLink);\n",
              "        }\n",
              "      </script>\n",
              "    </div>\n",
              "  </div>\n",
              "  "
            ]
          },
          "metadata": {},
          "execution_count": 2
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "D6AoKzPCKY9c"
      },
      "source": [
        "Esse foi um dos projetos mais extensos que já fiz e por isso vou mostrar como ficou dividido, assim você (leitor) pode ir direto até a parte que lhe interessa.\n",
        "\n",
        "O projeto ficou dividido da seguinte forma: \n",
        "\n",
        "\n",
        "*   Análise Exploratória\n",
        "\n",
        "*   Limpeza dos Dados\n",
        "\n",
        "*   Processamento dos Dados com Label Encoder e Get Dummies\n",
        "\n",
        "*   Preparação dos Dados\n",
        "\n",
        "  * Padronização e Balanceamento dos dados\n",
        "\n",
        "      * Balanceamento com Random UnderSampling \n",
        "\n",
        "      * Balanceamento com OverSampling SMOTE\n",
        "\n",
        "      * Balanceamento com OverSampling ADASYN\n",
        "\n",
        "  * Avaliando Modelos com Cross-Validation (Dados Balanceados e Padronizados)\n",
        "\n",
        "  * Balanceamento RUS\n",
        "      \n",
        "      * StandardScaler\n",
        "      \n",
        "      * RobustScaler\n",
        "\n",
        "      * Quantile Transformer\n",
        "\n",
        "      * Quantile Transformer-Uniform\n",
        "\n",
        "      * PowerTransformer\n",
        "\n",
        "  * Balanceamento SMOTE\n",
        "\n",
        "  * Balanceamento ADASYN\n",
        "\n",
        "  * Definindo Modelos para o Balanceamento RUS\n",
        "      \n",
        "      * LinearDiscriminantAnalysis\n",
        "\n",
        "      * SVC\n",
        "\n",
        "      * XGBoost\n",
        "\n",
        "  * Método Ensemble para os Modelos do Balanceamento RUS\n",
        "      \n",
        "      * Voting Classifier (RUS)\n",
        "\n",
        "  * Definindo Modelos para o Balanceamento SMOTE\n",
        "      \n",
        "      * Random Forest\n",
        "\n",
        "      * XGBoost\n",
        "\n",
        "      * LinearDiscriminantAnalysis\n",
        "\n",
        "  * Método Ensemble para os Modelos do Balanceamento SMOTE\n",
        "      \n",
        "      * Voting Classifier (SMOTE)\n",
        "\n",
        "  * Definindo Modelos para o Balanceamento ADASYN\n",
        "      \n",
        "      * Random Forest\n",
        "\n",
        "      * XGBoost\n",
        "\n",
        "      * LinearDiscriminantAnalysis\n",
        "\n",
        "  * Método Ensemble para os Modelos do Balanceamento ADASYN\n",
        "      \n",
        "      * Voting Classifier (ADASYN)\n",
        "\n",
        "* Conclusão\n",
        "\n",
        "Primeiro, vamos aprender o que é churn, de forma geral o churn é o número de clientes que cancelam um serviço em um determinado período de tempo, afim de solucionar ou amenizar esse número, as empresas podem utilizar modelos de machine learning para prever se um cliente pretende ou não cancelar o serviço.\n",
        "\n",
        "Os modelos de machine learning presentes no projeto, têm o objetivo de auxiliar empresas a lidar com o churn. Para isso, foram construídos modelos de aprendizagem supervisionada, como métrica para validação usamos o **recall** e para sabermos a perfomance minima esperada, podemos usar o **AUC (Área sob a Curva)**, sendo este necessário um score de **ao menos 0.70**.\n",
        "\n",
        "\n",
        "\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "542MqNwadeg3"
      },
      "source": [
        "# Análise Exploratória\n",
        "\n",
        "Vamos começar nossa análise imprimindo as dimensões do DataFrame, após isso checar o nome das colunas e seus respectivos tipos de dados e também realizar uma checagem de quantos valores **únicos** temos por coluna. Sabendo a quantidade de valores únicos por coluna, podemos identificar colunas que podem ou não ser categóricas."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "c1uUEjKOctBk",
        "outputId": "7dfe9623-868d-49ec-c421-7ffe08f7f0e8"
      },
      "source": [
        "# Checar as dimensões do DataFrame\n",
        "print(f'O DataFrame possui {df.shape[0]} entradas e {df.shape[1]} colunas.')\n",
        "\n",
        "# Checar o nome das colunas\n",
        "print(f'\\nNome das Colunas:\\n{df.columns.values}')\n",
        "\n",
        "# Checar o tipo de dados (por coluna)\n",
        "print(f'\\nTipos de dados (por coluna)\\n{df.dtypes}')\n",
        "\n",
        "# Checar os valores únicos por coluna\n",
        "print(f'\\nValores únicos (por coluna)\\n{df.nunique()}')\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "O DataFrame possui 7043 entradas e 21 colunas.\n",
            "\n",
            "Nome das Colunas:\n",
            "['customerID' 'gender' 'SeniorCitizen' 'Partner' 'Dependents' 'tenure'\n",
            " 'PhoneService' 'MultipleLines' 'InternetService' 'OnlineSecurity'\n",
            " 'OnlineBackup' 'DeviceProtection' 'TechSupport' 'StreamingTV'\n",
            " 'StreamingMovies' 'Contract' 'PaperlessBilling' 'PaymentMethod'\n",
            " 'MonthlyCharges' 'TotalCharges' 'Churn']\n",
            "\n",
            "Tipos de dados (por coluna)\n",
            "customerID           object\n",
            "gender               object\n",
            "SeniorCitizen         int64\n",
            "Partner              object\n",
            "Dependents           object\n",
            "tenure                int64\n",
            "PhoneService         object\n",
            "MultipleLines        object\n",
            "InternetService      object\n",
            "OnlineSecurity       object\n",
            "OnlineBackup         object\n",
            "DeviceProtection     object\n",
            "TechSupport          object\n",
            "StreamingTV          object\n",
            "StreamingMovies      object\n",
            "Contract             object\n",
            "PaperlessBilling     object\n",
            "PaymentMethod        object\n",
            "MonthlyCharges      float64\n",
            "TotalCharges         object\n",
            "Churn                object\n",
            "dtype: object\n",
            "\n",
            "Valores únicos (por coluna)\n",
            "customerID          7043\n",
            "gender                 2\n",
            "SeniorCitizen          2\n",
            "Partner                2\n",
            "Dependents             2\n",
            "tenure                73\n",
            "PhoneService           2\n",
            "MultipleLines          3\n",
            "InternetService        3\n",
            "OnlineSecurity         3\n",
            "OnlineBackup           3\n",
            "DeviceProtection       3\n",
            "TechSupport            3\n",
            "StreamingTV            3\n",
            "StreamingMovies        3\n",
            "Contract               3\n",
            "PaperlessBilling       2\n",
            "PaymentMethod          4\n",
            "MonthlyCharges      1585\n",
            "TotalCharges        6531\n",
            "Churn                  2\n",
            "dtype: int64\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "HbvGRERmTlv6"
      },
      "source": [
        "Para uma análise bem feita é necessário também checar a quantidade de dados ausentes."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "gZK-BYJsc3er",
        "outputId": "bab24d19-317b-4891-b100-d40f66c644cc"
      },
      "source": [
        "# Checar a quantidade de dados ausentes\n",
        "print(f'Porcentagem de dados ausentes:\\t\\n{df.isnull().sum() / df.shape[0]}')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Porcentagem de dados ausentes:\t\n",
            "customerID          0.0\n",
            "gender              0.0\n",
            "SeniorCitizen       0.0\n",
            "Partner             0.0\n",
            "Dependents          0.0\n",
            "tenure              0.0\n",
            "PhoneService        0.0\n",
            "MultipleLines       0.0\n",
            "InternetService     0.0\n",
            "OnlineSecurity      0.0\n",
            "OnlineBackup        0.0\n",
            "DeviceProtection    0.0\n",
            "TechSupport         0.0\n",
            "StreamingTV         0.0\n",
            "StreamingMovies     0.0\n",
            "Contract            0.0\n",
            "PaperlessBilling    0.0\n",
            "PaymentMethod       0.0\n",
            "MonthlyCharges      0.0\n",
            "TotalCharges        0.0\n",
            "Churn               0.0\n",
            "dtype: float64\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "mMFzQT1LRsJw"
      },
      "source": [
        "Note que existem algumas observações a serem feitas:\n",
        "\n",
        "* A coluna `customerID` é insignificante para a análise, uma vez que apenas nos fornece o identificador de cada cliente.\n",
        "\n",
        "* A variável alvo está presente na coluna `Churn`, ela indica se o cliente cancelou ou não o serviço.\n",
        "\n",
        "* A coluna `TotalCharges` assim como a coluna `MonthlyCharges`, apresenta valores flutuantes, mas está como string uma vez que seu tipo de dado deveria ser `float`.\n",
        "\n",
        "* Diversas colunas vão precisar de um trabalho de codificação, transformando-as em valores numéricos.\n",
        "\n",
        "* Apesar do nosso código retornar que não há valores ausentes, pode ser que algumas strings representem valores ausentes.\n",
        "\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "d-l7J6fCLEFq"
      },
      "source": [
        "# Limpeza dos Dados"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "kGIfbsbnUDTV"
      },
      "source": [
        "Anteriormente foi dito que a coluna `TotalCharges` deveria ser apresentada como `float`, mas apresentava uma string. \n",
        "\n",
        "Devido a isso, fui obrigado a usar um método mais complexo para converter a coluna, pois ao usar \n",
        "`df[TotalCharges].astype('float32')` é retornado um erro.\n",
        "\n",
        "O erro em questão acontece devido aos dados ausentes terem sido preenchidos com `\"\"`. Uma vez que tentamos converter `\"\"` para float, o interpretador nos retorna um erro.\n",
        "\n",
        "Para contornar esse erro e converter de forma adequada foi criada uma função para lidar com o erro. Consequentemente, foi possível identificar valores ausentes, que logo foram preenchidos com o valor da mediana da coluna.\n",
        "\n",
        "Também foi excluída a coluna `customerID`, uma vez que é desnecessária para o projeto."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "lP2aldTPZGW-",
        "outputId": "0586d38e-bab2-4552-ec23-bfe5ae654a96"
      },
      "source": [
        "def str_to_float(entrada):\n",
        "  \"\"\"\n",
        "  Converte um objeto do tipo string em float.\n",
        "\n",
        "   Caso seja possível a string sera convertida em float\n",
        "   \n",
        "   Se a conversão no for permitida, a string é convertida em um valor NaN\n",
        "  \"\"\"\n",
        "  \n",
        "  try:\n",
        "    return float(entrada)\n",
        "  except ValueError:\n",
        "    return np.nan\n",
        "\n",
        "# Criar cópia do DataFrame\n",
        "df_clean = df.copy()\n",
        "\n",
        "# Excluir a coluna customerID\n",
        "df_clean.drop('customerID', axis=1, inplace=True)\n",
        "\n",
        "# Aplicando a função de conversão \n",
        "df_clean['TotalCharges'] = df_clean['TotalCharges'].apply(str_to_float)\n",
        "\n",
        "# Mostrar a quantidade de dados ausentes antes do preenchimento\n",
        "print('Dados ausentes em \"TotalCharges\" antes de preencher os dados: {}'.format(df_clean['TotalCharges'].isnull().sum()))\n",
        "\n",
        "# Preencher os dados ausentes com o valor da mediana da coluna\n",
        "total_charges_median = df_clean.TotalCharges.median()\n",
        "df_clean['TotalCharges'].fillna(total_charges_median, inplace=True)"
      ],
      "execution_count": 3,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Dados ausentes em \"TotalCharges\" antes de preencher os dados: 11\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "0iRodTXIc6Ze"
      },
      "source": [
        "Apesar de o DataFrame ser majoritariamente de variáveis categóricas, possuímos algumas variáves numéricas que com o método `describe` é possível notar que não há a presença de outliers."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 300
        },
        "id": "HLtoppTQc7db",
        "outputId": "3445c55e-22da-4df5-af49-064cde4981dd"
      },
      "source": [
        "# verificando dados estatisticos\n",
        "df_clean.describe()"
      ],
      "execution_count": 7,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "       SeniorCitizen       tenure  MonthlyCharges  TotalCharges\n",
              "count    7043.000000  7043.000000     7043.000000   7043.000000\n",
              "mean        0.162147    32.371149       64.761692   2281.916928\n",
              "std         0.368612    24.559481       30.090047   2265.270398\n",
              "min         0.000000     0.000000       18.250000     18.800000\n",
              "25%         0.000000     9.000000       35.500000    402.225000\n",
              "50%         0.000000    29.000000       70.350000   1397.475000\n",
              "75%         0.000000    55.000000       89.850000   3786.600000\n",
              "max         1.000000    72.000000      118.750000   8684.800000"
            ],
            "text/html": [
              "\n",
              "  <div id=\"df-93042da4-9ba6-4875-a9c7-6317237ded29\">\n",
              "    <div class=\"colab-df-container\">\n",
              "      <div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>SeniorCitizen</th>\n",
              "      <th>tenure</th>\n",
              "      <th>MonthlyCharges</th>\n",
              "      <th>TotalCharges</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>count</th>\n",
              "      <td>7043.000000</td>\n",
              "      <td>7043.000000</td>\n",
              "      <td>7043.000000</td>\n",
              "      <td>7043.000000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>mean</th>\n",
              "      <td>0.162147</td>\n",
              "      <td>32.371149</td>\n",
              "      <td>64.761692</td>\n",
              "      <td>2281.916928</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>std</th>\n",
              "      <td>0.368612</td>\n",
              "      <td>24.559481</td>\n",
              "      <td>30.090047</td>\n",
              "      <td>2265.270398</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>min</th>\n",
              "      <td>0.000000</td>\n",
              "      <td>0.000000</td>\n",
              "      <td>18.250000</td>\n",
              "      <td>18.800000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>25%</th>\n",
              "      <td>0.000000</td>\n",
              "      <td>9.000000</td>\n",
              "      <td>35.500000</td>\n",
              "      <td>402.225000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>50%</th>\n",
              "      <td>0.000000</td>\n",
              "      <td>29.000000</td>\n",
              "      <td>70.350000</td>\n",
              "      <td>1397.475000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>75%</th>\n",
              "      <td>0.000000</td>\n",
              "      <td>55.000000</td>\n",
              "      <td>89.850000</td>\n",
              "      <td>3786.600000</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>max</th>\n",
              "      <td>1.000000</td>\n",
              "      <td>72.000000</td>\n",
              "      <td>118.750000</td>\n",
              "      <td>8684.800000</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-93042da4-9ba6-4875-a9c7-6317237ded29')\"\n",
              "              title=\"Convert this dataframe to an interactive table.\"\n",
              "              style=\"display:none;\">\n",
              "        \n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "       width=\"24px\">\n",
              "    <path d=\"M0 0h24v24H0V0z\" fill=\"none\"/>\n",
              "    <path d=\"M18.56 5.44l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94zm-11 1L8.5 8.5l.94-2.06 2.06-.94-2.06-.94L8.5 2.5l-.94 2.06-2.06.94zm10 10l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94z\"/><path d=\"M17.41 7.96l-1.37-1.37c-.4-.4-.92-.59-1.43-.59-.52 0-1.04.2-1.43.59L10.3 9.45l-7.72 7.72c-.78.78-.78 2.05 0 2.83L4 21.41c.39.39.9.59 1.41.59.51 0 1.02-.2 1.41-.59l7.78-7.78 2.81-2.81c.8-.78.8-2.07 0-2.86zM5.41 20L4 18.59l7.72-7.72 1.47 1.35L5.41 20z\"/>\n",
              "  </svg>\n",
              "      </button>\n",
              "      \n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      flex-wrap:wrap;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "      <script>\n",
              "        const buttonEl =\n",
              "          document.querySelector('#df-93042da4-9ba6-4875-a9c7-6317237ded29 button.colab-df-convert');\n",
              "        buttonEl.style.display =\n",
              "          google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "        async function convertToInteractive(key) {\n",
              "          const element = document.querySelector('#df-93042da4-9ba6-4875-a9c7-6317237ded29');\n",
              "          const dataTable =\n",
              "            await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                     [key], {});\n",
              "          if (!dataTable) return;\n",
              "\n",
              "          const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "            '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "            + ' to learn more about interactive tables.';\n",
              "          element.innerHTML = '';\n",
              "          dataTable['output_type'] = 'display_data';\n",
              "          await google.colab.output.renderOutput(dataTable, element);\n",
              "          const docLink = document.createElement('div');\n",
              "          docLink.innerHTML = docLinkHtml;\n",
              "          element.appendChild(docLink);\n",
              "        }\n",
              "      </script>\n",
              "    </div>\n",
              "  </div>\n",
              "  "
            ]
          },
          "metadata": {},
          "execution_count": 7
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "_ssJl1M1XDsQ"
      },
      "source": [
        "É possível confirmar a falta de outliers quando analisamos os boxplots das colunas `MonthlyCharges` e `TotalCharges`, uma vez que mesmo sem definirmos limites não foi encontrado nenhum plot fora dos limites."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 298
        },
        "id": "HuJ2WWVlc_j9",
        "outputId": "6619e676-d5ca-4fff-e35d-0006553c8e9e"
      },
      "source": [
        "# Plotar gráficos boxplot para verificar a presença de outliers\n",
        "fig , ax = plt.subplots(nrows=1, ncols=2, figsize=(15,5))\n",
        "sns.boxplot(df_clean['MonthlyCharges'], ax=ax[0])\n",
        "sns.boxplot(df_clean['TotalCharges'], ax=ax[1])\n",
        "\n",
        "plt.tight_layout()\n",
        "\n",
        "plt.savefig('boxplot-outliers-churn.jpeg')"
      ],
      "execution_count": 4,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 1080x360 with 2 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ifnyw_r3dFse"
      },
      "source": [
        "Para termos uma noção de quanto tempo um cliente permanece no serviço, podemos verificar essa \"Fidelidade\" plotando um gráfico de barras da varáivel `tenure`, que provavelmente usa como unidade de tempo o \"mês\"."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 298
        },
        "id": "uriJi5KgdDn2",
        "outputId": "1ef1c166-f003-402f-c948-52a788cfdd5f"
      },
      "source": [
        "# Plotar o gráfico\n",
        "fig, ax = plt.subplots(figsize=(15, 5))\n",
        "sns.countplot(df_clean['tenure'], data=df_clean, palette='Spectral', ax=ax)\n",
        "\n",
        "# Definir legendas para o gráfico\n",
        "ax.set_title('Tempo de assinatura dos Clientes', fontsize=14\n",
        ")\n",
        "ax.set_ylabel('Quantidade de Clientes', fontsize=14)\n",
        "ax.set_xlabel('Tempo (Meses)', fontsize=14)\n",
        "\n",
        "plt.tight_layout()\n",
        "\n",
        "plt.savefig('Tempo_meses_churn.jpeg', dpi=300)"
      ],
      "execution_count": 16,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 1080x360 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "uE2cSsYdX-IH"
      },
      "source": [
        "Ao lidar com variáveis categóricas, se torna necessário identificar os valores únicos. \n",
        "\n",
        "Abaixo podemos ver que existem valores que podem ser unificados, como por exemplo:\n",
        "* `No`\n",
        "\n",
        "* `No phone service`\n",
        "\n",
        "* `No internet service`.\n",
        "\n",
        "Para isso, irei apenas definir os valores como No, assim  não teremos valores que representam a mesma coisa."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "x6TH6SNx2bsm",
        "outputId": "604d753d-c114-4611-ed9d-91e0a7634f30"
      },
      "source": [
        "# verificar valores unicos para variaveis categoricas\n",
        "np.unique(df_clean.select_dtypes('object').values)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array(['Bank transfer (automatic)', 'Credit card (automatic)', 'DSL',\n",
              "       'Electronic check', 'Female', 'Fiber optic', 'Mailed check',\n",
              "       'Male', 'Month-to-month', 'No', 'No internet service',\n",
              "       'No phone service', 'One year', 'Two year', 'Yes'], dtype=object)"
            ]
          },
          "metadata": {},
          "execution_count": 9
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-tC4q04i5t5U"
      },
      "source": [
        "# Substituindo \"No phone service\" e \"No internet service\" por \"No\"\n",
        "df_clean.replace({'No internet service': 'No'}, regex=True, inplace=True)\n",
        "df_clean.replace({'No phone service': 'No'}, regex=True, inplace=True)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "zx6F-cL47ZO8",
        "outputId": "6531c45c-1760-4196-8be7-047d3c25e55a"
      },
      "source": [
        "# Checando a mudança nos valores únicos\n",
        "np.unique(df_clean.select_dtypes('object').values)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array(['Bank transfer (automatic)', 'Credit card (automatic)', 'DSL',\n",
              "       'Electronic check', 'Female', 'Fiber optic', 'Mailed check',\n",
              "       'Male', 'Month-to-month', 'No', 'One year', 'Two year', 'Yes'],\n",
              "      dtype=object)"
            ]
          },
          "metadata": {},
          "execution_count": 11
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "M40z6-04LtJA"
      },
      "source": [
        "Para concluir a análise exploratória, vou verificar o balanceamento das classes.\n",
        "\n",
        "É notável a diferença entre as classes, uma vez que a classe `No` ocorre com frequência superior à classe `Yes`. Para obtermos melhores resultados com nossos modelos de Machine Learning é necessário fazer o balanceamento das classes, isso será feito ao decorrer do projeto."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 401
        },
        "id": "zpKt2y16dGrp",
        "outputId": "55757a2c-4162-4cbc-ba10-5bbbac0ce822"
      },
      "source": [
        "# Plotar gráfico das classes\n",
        "print(df_clean.Churn.value_counts())\n",
        "print('\\nChurns representam {:.4f}% do dataset.\\n'.format((df_clean[df_clean.Churn == \"Yes\"].shape[0] / df.shape[0]) * 100))\n",
        "\n",
        "sns.countplot(df_clean.Churn)\n",
        "plt.tight_layout()\n",
        "\n",
        "plt.savefig('balanceamento_churn.jpeg', dpi=300)"
      ],
      "execution_count": 17,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "No     5174\n",
            "Yes    1869\n",
            "Name: Churn, dtype: int64\n",
            "\n",
            "Churns representam 26.5370% do dataset.\n",
            "\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "DNUtlsrFdQUB"
      },
      "source": [
        "# Processamento dos Dados com Label Encoder e Get Dummies\n",
        "\n",
        "Nesta etapa é necessário um pré-processamento dos dados. As colunas binárias (que possuem apenas dois valores únicos), serão processadas usando o `LabelEncoder`, incluindo a variável alvo `Churn`.\n",
        "\n",
        "Em seguida, as variáveis serão separadas em categóricas e numéricas, onde as variáveis categóricas serão tratadas com o `getDummies`, assim poderão ser utilizadas em todos os modelos."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 261
        },
        "id": "cMbEO_bi-8tA",
        "outputId": "5a1da194-5d21-494b-9567-994c0c0c7606"
      },
      "source": [
        "# localizando as variaveis e seus tipos\n",
        "variaveis_binarias = df_clean.nunique()[df_clean.nunique() == 2].keys().tolist()\n",
        "variaveis_numericas = [col for col in df_clean.select_dtypes(['int', 'float']).columns.tolist() if col not in variaveis_binarias]\n",
        "variaveis_categoricas = [col for col in df_clean.columns.tolist() if col not in variaveis_binarias + variaveis_numericas]\n",
        "\n",
        "df_clean_proc = df_clean.copy()\n",
        "\n",
        "# Processando os dados com o Label Encoder\n",
        "le = LabelEncoder()\n",
        "for i in variaveis_binarias:\n",
        "  df_clean_proc[i] = le.fit_transform(df_clean_proc[i])\n",
        "\n",
        "# Processandos os dados com o Get Dummies\n",
        "df_clean_proc = pd.get_dummies(df_clean_proc, columns=variaveis_categoricas)\n",
        "\n",
        "df_clean_proc.head()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>gender</th>\n",
              "      <th>SeniorCitizen</th>\n",
              "      <th>Partner</th>\n",
              "      <th>Dependents</th>\n",
              "      <th>tenure</th>\n",
              "      <th>PhoneService</th>\n",
              "      <th>MultipleLines</th>\n",
              "      <th>OnlineSecurity</th>\n",
              "      <th>OnlineBackup</th>\n",
              "      <th>DeviceProtection</th>\n",
              "      <th>TechSupport</th>\n",
              "      <th>StreamingTV</th>\n",
              "      <th>StreamingMovies</th>\n",
              "      <th>PaperlessBilling</th>\n",
              "      <th>MonthlyCharges</th>\n",
              "      <th>TotalCharges</th>\n",
              "      <th>Churn</th>\n",
              "      <th>InternetService_DSL</th>\n",
              "      <th>InternetService_Fiber optic</th>\n",
              "      <th>InternetService_No</th>\n",
              "      <th>Contract_Month-to-month</th>\n",
              "      <th>Contract_One year</th>\n",
              "      <th>Contract_Two year</th>\n",
              "      <th>PaymentMethod_Bank transfer (automatic)</th>\n",
              "      <th>PaymentMethod_Credit card (automatic)</th>\n",
              "      <th>PaymentMethod_Electronic check</th>\n",
              "      <th>PaymentMethod_Mailed check</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "      <td>29.85</td>\n",
              "      <td>29.85</td>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>34</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>56.95</td>\n",
              "      <td>1889.50</td>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>2</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "      <td>53.85</td>\n",
              "      <td>108.15</td>\n",
              "      <td>1</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>45</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>42.30</td>\n",
              "      <td>1840.75</td>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>2</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "      <td>70.70</td>\n",
              "      <td>151.65</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "      <td>0</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "   gender  ...  PaymentMethod_Mailed check\n",
              "0       0  ...                           0\n",
              "1       1  ...                           1\n",
              "2       1  ...                           1\n",
              "3       1  ...                           0\n",
              "4       0  ...                           0\n",
              "\n",
              "[5 rows x 27 columns]"
            ]
          },
          "metadata": {},
          "execution_count": 14
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "SOAItkZWdIsp"
      },
      "source": [
        "É possível notar que mesmo sem fazer um balanceamento dos dados e uma padronização, temos uma boa correlação entre algumas variáveis, que possivelmente ficará ainda melhor após estes processos."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 583
        },
        "id": "ExoGFp24gt6m",
        "outputId": "858f9874-2b90-437a-f918-d347e8d06def"
      },
      "source": [
        "# Plotar matriz de correlação\n",
        "fig, ax = plt.subplots(figsize=(8,6))\n",
        "sns.heatmap(df_clean_proc.corr())\n",
        "\n",
        "plt.show();"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 576x432 with 2 Axes>"
            ]
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "xUm79j9bceyL"
      },
      "source": [
        "# Preparação dos Dados\n",
        "\n",
        "Nesta seção, vou realizar a separação do nosso DF em treino e teste e identificar a nossa variável alvo. Além de também efetuar a padronização dos dados, será feita uma avaliação dos modelos por meio do Cross-Validation com os dados desbalanceados e posteriormente com os dados balanceados e padronizados."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "bi7dnGylMybY"
      },
      "source": [
        "**Separar os dados entre feature matrix e target vector**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "6qHJrOBIrpjv"
      },
      "source": [
        "# separar os dados enter feature matrix e target vector\n",
        "X = df_clean_proc.drop('Churn', axis=1)\n",
        "y = df_clean_proc['Churn']\n",
        "\n",
        "# Dividir o dataset entre treino e teste\n",
        "X_train, X_test, y_train, y_test = train_test_split(X, y, stratify=y, shuffle=True)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "2-J-eQVsNAtR"
      },
      "source": [
        "**Avaliando Modelos com Cross-Validation**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "hkGov1yv2cpP"
      },
      "source": [
        "# Função de validação dos modelos\n",
        "def val_model(X, y,clf, quite=False):\n",
        "\n",
        "  X = np.array(X)\n",
        "  y = np.array(y)\n",
        "\n",
        "# Definindo a padronização antes da validação\n",
        "  pipeline1 = make_pipeline(StandardScaler(), clf)\n",
        "  pipeline2 = make_pipeline(RobustScaler(), clf)\n",
        "  pipeline3 = make_pipeline(QuantileTransformer(), clf)\n",
        "  pipeline4 = make_pipeline(QuantileTransformer(output_distribution='normal'), clf)\n",
        "  pipeline5 = make_pipeline(PowerTransformer(method='yeo-johnson'), clf)\n",
        "  scores1 = cross_val_score(pipeline1, X, y, scoring='recall')\n",
        "  scores2 = cross_val_score(pipeline2, X, y, scoring='recall')\n",
        "  scores3 = cross_val_score(pipeline3, X, y, scoring='recall')\n",
        "  scores4 = cross_val_score(pipeline4, X, y, scoring='recall')\n",
        "  scores5 = cross_val_score(pipeline5, X, y, scoring='recall')\n",
        "\n",
        "  if quite == False:\n",
        "    print('Recall StandardScaler: {:.4f} (+/- {:.4f})'.format(scores1.mean(), scores1.std() * 2))\n",
        "    print('Recall RobustScaler: {:.4f} (+/- {:.4f})'.format(scores2.mean(), scores2.std() * 2))\n",
        "    print('Recall QuantileTransformer: {:.4f} (+/- {:.4f})'.format(scores3.mean(), scores3.std() * 2))\n",
        "    print('Recall QuantileTransformer-N: {:.4f} (+/- {:.4f})'.format(scores4.mean(), scores4.std() * 2))\n",
        "    print('Recall PowerTransformer: {:.4f} (+/- {:.4f})'.format(scores5.mean(), scores5.std() * 2))\n",
        "  return scores1.mean()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "SeRpyCqauthY"
      },
      "source": [
        "# Importar Modelos a serem avaliados\n",
        "rf = RandomForestClassifier()\n",
        "lr = LogisticRegression()\n",
        "xgb = XGBClassifier()\n",
        "lda = LinearDiscriminantAnalysis()\n",
        "svc = SVC()\n",
        "sgd = SGDClassifier()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "D_0gOzTL5SfL",
        "outputId": "8505422b-3977-40cd-96ef-aa9293fdce4e"
      },
      "source": [
        "# printar o desempenho dos modelos com os dados padronizados\n",
        "print('Cross-validation RF:')\n",
        "score_teste1 = val_model(X_train, y_train, rf)\n",
        "print('\\nCross-validation LR:')\n",
        "score_teste2 = val_model(X_train, y_train, lr)\n",
        "print('\\nCross-validation XGB:')\n",
        "score_teste3 = val_model(X_train, y_train, xgb)\n",
        "print('\\nCross-validation LDA:')\n",
        "score_teste4 = val_model(X_train, y_train, lda)\n",
        "print('\\nCross-validation SVC:')\n",
        "score_teste5 = val_model(X_train, y_train, svc)\n",
        "print('\\nCross-validation SGD:')\n",
        "score_teste6 = val_model(X_train, y_train, sgd)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Cross-validation RF:\n",
            "Recall StandardScaler: 0.4950 (+/- 0.0570)\n",
            "Recall RobustScaler: 0.4808 (+/- 0.0655)\n",
            "Recall QuantileTransformer: 0.4843 (+/- 0.0615)\n",
            "Recall QuantileTransformer-N: 0.4929 (+/- 0.0475)\n",
            "Recall PowerTransformer: 0.4829 (+/- 0.0465)\n",
            "\n",
            "Cross-validation LR:\n",
            "Recall StandardScaler: 0.5635 (+/- 0.0726)\n",
            "Recall RobustScaler: 0.5606 (+/- 0.0706)\n",
            "Recall QuantileTransformer: 0.5471 (+/- 0.0702)\n",
            "Recall QuantileTransformer-N: 0.5571 (+/- 0.0617)\n",
            "Recall PowerTransformer: 0.5385 (+/- 0.0564)\n",
            "\n",
            "Cross-validation XGB:\n",
            "Recall StandardScaler: 0.5421 (+/- 0.0484)\n",
            "Recall RobustScaler: 0.5421 (+/- 0.0484)\n",
            "Recall QuantileTransformer: 0.5421 (+/- 0.0484)\n",
            "Recall QuantileTransformer-N: 0.5421 (+/- 0.0484)\n",
            "Recall PowerTransformer: 0.5421 (+/- 0.0484)\n",
            "\n",
            "Cross-validation LDA:\n",
            "Recall StandardScaler: 0.5692 (+/- 0.0584)\n",
            "Recall RobustScaler: 0.5692 (+/- 0.0584)\n",
            "Recall QuantileTransformer: 0.5385 (+/- 0.0618)\n",
            "Recall QuantileTransformer-N: 0.5706 (+/- 0.0458)\n",
            "Recall PowerTransformer: 0.5414 (+/- 0.0476)\n",
            "\n",
            "Cross-validation SVC:\n",
            "Recall StandardScaler: 0.5228 (+/- 0.0740)\n",
            "Recall RobustScaler: 0.5129 (+/- 0.0658)\n",
            "Recall QuantileTransformer: 0.5036 (+/- 0.0712)\n",
            "Recall QuantileTransformer-N: 0.4907 (+/- 0.0572)\n",
            "Recall PowerTransformer: 0.5043 (+/- 0.0733)\n",
            "\n",
            "Cross-validation SGD:\n",
            "Recall StandardScaler: 0.5584 (+/- 0.1122)\n",
            "Recall RobustScaler: 0.5606 (+/- 0.1121)\n",
            "Recall QuantileTransformer: 0.5584 (+/- 0.2496)\n",
            "Recall QuantileTransformer-N: 0.6333 (+/- 0.2218)\n",
            "Recall PowerTransformer: 0.4680 (+/- 0.1309)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "I1WeCAmlN1BJ"
      },
      "source": [
        "## Padronização e Balanceamento dos Dados\n",
        "\n",
        "Para a padronização, serão usados 5 métodos, sendo assim poderemos ter uma melhor comparação de como os modelos se saem com esses dados em cada tipo de padronização. Após a padronização, os dados padronizados serão combinados com diferentes métodos de balanceamento de dados."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "QwQBsTArpZNB"
      },
      "source": [
        "# importar metodos para realização do Feature Scaling\n",
        "scaler = StandardScaler()\n",
        "Rob_scaler = RobustScaler()\n",
        "qtr = QuantileTransformer()\n",
        "qtr_n = QuantileTransformer(output_distribution='normal')\n",
        "pwr = PowerTransformer()\n",
        "\n",
        "# Padronizando os dados de treino\n",
        "X_train_scaled = scaler.fit_transform(X_train)\n",
        "X_train_Rscaled = Rob_scaler.fit_transform(X_train)\n",
        "X_train_qtr = qtr.fit_transform(X_train)\n",
        "X_train_qtr_n = qtr_n.fit_transform(X_train)\n",
        "X_train_pwr = pwr.fit_transform(X_train)\n",
        "\n",
        "# Balanceamento RUS\n",
        "rus = RandomUnderSampler()\n",
        "X_rus_scaled, y_train_rus_scaled = rus.fit_resample(X_train_scaled, y_train)\n",
        "X_rus_Rscaled, y_train_rus_Rscaled = rus.fit_resample(X_train_Rscaled, y_train)\n",
        "X_rus_qtr, y_train_rus_qtr = rus.fit_resample(X_train_qtr, y_train)\n",
        "X_rus_qtr_n, y_train_rus_qtr_n = rus.fit_resample(X_train_qtr_n, y_train)\n",
        "X_rus_pwr, y_train_rus_pwr = rus.fit_resample(X_train_pwr, y_train)\n",
        "\n",
        "# Balanceamento SMOTE\n",
        "smo = SMOTE()\n",
        "X_smo_scaled, y_train_smo_scaled = smo.fit_resample(X_train_scaled, y_train)\n",
        "X_smo_Rscaled, y_train_smo_Rscaled = smo.fit_resample(X_train_Rscaled, y_train)\n",
        "X_smo_qtr, y_train_smo_qtr = smo.fit_resample(X_train_qtr, y_train)\n",
        "X_smo_qtr_n, y_train_smo_qtr_n = smo.fit_resample(X_train_qtr_n, y_train)\n",
        "X_smo_pwr, y_train_smo_pwr = smo.fit_resample(X_train_pwr, y_train)\n",
        "\n",
        "# Balanceamento ADASYN\n",
        "ada = ADASYN()\n",
        "X_ada_scaled, y_train_ada_scaled = ada.fit_resample(X_train_scaled, y_train)\n",
        "X_ada_Rscaled, y_train_ada_Rscaled = ada.fit_resample(X_train_Rscaled, y_train)\n",
        "X_ada_qtr, y_train_ada_qtr = ada.fit_resample(X_train_qtr, y_train)\n",
        "X_ada_qtr_n, y_train_ada_qtr_n = ada.fit_resample(X_train_qtr_n, y_train)\n",
        "X_ada_pwr, y_train_ada_pwr = ada.fit_resample(X_train_pwr, y_train)\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "8pQikbUqN_mb"
      },
      "source": [
        "### Balanceamento com Random UnderSampling\n",
        "\n",
        "Aqui eu começo a realizar o balanceamento da variável `Churn`, usando RandomUnderSampling, assim serão excluídas boa parte das entradas da classe majoritária e poucas entradas da classe minoritária.\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 348
        },
        "id": "-sA-_QqCsi67",
        "outputId": "b5ad5760-906a-49ff-d46d-16b505b70247"
      },
      "source": [
        "# Checando o balanceamento das classes\n",
        "print(pd.Series(y_train_rus_scaled).value_counts())\n",
        "\n",
        "fig, ax = plt.subplots() \n",
        "sns.countplot(y_train_rus_scaled)\n",
        "\n",
        "plt.tight_layout()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "1    1402\n",
            "0    1402\n",
            "dtype: int64\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "i2bBSssdOJB5"
      },
      "source": [
        "### Balanceamento com OverSampling SMOTE\n",
        "\n",
        "Neste segundo balanceamento, SMOTE,são adicionadas entradas para a classe minoritária, com caracteristicas semelhantes as já existentes.\n",
        "\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 348
        },
        "id": "p9qxKPtYtBe1",
        "outputId": "6933907b-ef4f-463a-8226-11b37ad80af8"
      },
      "source": [
        "# Checando o balanceamento das classes\n",
        "print(pd.Series(y_train_smo_scaled).value_counts())\n",
        "\n",
        "fig, ax = plt.subplots() \n",
        "sns.countplot(y_train_smo_scaled)\n",
        "\n",
        "plt.tight_layout()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "1    3880\n",
            "0    3880\n",
            "dtype: int64\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "r6ofUtz5OPXD"
      },
      "source": [
        "### Balanceamento com OverSampling ADASYN\n",
        "\n",
        "O terceiro balanceamento, ADASYN, também adiciona entradas mas tenta fazer uma diferenciação das entradas já existentes, levando em consideração a complexidade dos dados de cada entrada."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 348
        },
        "id": "JNCJzp0otVdL",
        "outputId": "8255058f-3e90-44dd-c146-e3b9aeb84873"
      },
      "source": [
        "# Checando o balanceamento das classes\n",
        "print(pd.Series(y_train_ada_scaled).value_counts())\n",
        "\n",
        "fig, ax = plt.subplots() \n",
        "\n",
        "sns.countplot(y_train_ada_scaled)\n",
        "\n",
        "\n",
        "plt.tight_layout();"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "1    3892\n",
            "0    3880\n",
            "dtype: int64\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Q0Hnu5jHOUa-"
      },
      "source": [
        "## Avaliandos Modelos com Cross-Validation (Dados Balanceados e Padronizados)\n",
        "\n",
        "Como dito anteriormente, vou avaliar os modelos novamente, agora com os dados balanceados e padronizados, depois irei selecionar 3 modelos para cada tipo de balanceamento. Assim, posso usar o método ensemble para combinar os modelos e possivelmente obter um resultado mais constante e melhor."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FM2UNiSiMtUl"
      },
      "source": [
        "# Definindo funçao de validação com dados balanceados\n",
        "def val_model_balanced(X, y,clf, quite=False):\n",
        "\n",
        "  X = np.array(X)\n",
        "  y = np.array(y)\n",
        "\n",
        "  scores = cross_val_score(clf, X, y, scoring='recall')\n",
        "\n",
        "  if quite == False:\n",
        "    print('Recall: {:.4f} (+/- {:.4f})'.format(scores.mean(), scores.std() * 2))\n",
        "  return scores.mean()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Mw-qrGgNOh0S"
      },
      "source": [
        "## Balanceamento RUS com StandardScaler (scaled)\n",
        "As validações serão avaliadas pela métrica recall e serão dividas. Primeiramente, em balanceamentos e com subgrupos que depende do tipo de padronização utilizada."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "D5YCIw9hJlYt",
        "outputId": "dbb6240f-6fcf-4b2a-92bd-c69f66d47c91"
      },
      "source": [
        "# Cross-Validation com Balanceamento RUS e Standard Scaler\n",
        "print('Cross-validation RF:')\n",
        "score_teste1 = val_model_balanced(X_rus_scaled, y_train_rus_scaled, rf)\n",
        "print('\\nCross-validation LR:')\n",
        "score_teste2 = val_model_balanced(X_rus_scaled, y_train_rus_scaled, lr)\n",
        "print('\\nCross-validation XGB:')\n",
        "score_teste3 = val_model_balanced(X_rus_scaled, y_train_rus_scaled, xgb)\n",
        "print('\\nCross-validation LDA:')\n",
        "score_teste4 = val_model_balanced(X_rus_scaled, y_train_rus_scaled, lda)\n",
        "print('\\nCross-validation SVC:')\n",
        "score_teste5 = val_model_balanced(X_rus_scaled, y_train_rus_scaled, svc)\n",
        "print('\\nCross-validation SGD:')\n",
        "score_teste6 = val_model_balanced(X_rus_scaled, y_train_rus_scaled, sgd)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Cross-validation RF:\n",
            "Recall: 0.7825 (+/- 0.0253)\n",
            "\n",
            "Cross-validation LR:\n",
            "Recall: 0.8160 (+/- 0.0268)\n",
            "\n",
            "Cross-validation XGB:\n",
            "Recall: 0.8267 (+/- 0.0387)\n",
            "\n",
            "Cross-validation LDA:\n",
            "Recall: 0.8088 (+/- 0.0342)\n",
            "\n",
            "Cross-validation SVC:\n",
            "Recall: 0.8003 (+/- 0.0249)\n",
            "\n",
            "Cross-validation SGD:\n",
            "Recall: 0.7489 (+/- 0.0739)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "2CnkkDyMOr3s"
      },
      "source": [
        "### Balanceamento RUS com RobustScaler (Rscaled)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "_n_mcaepJmak",
        "outputId": "50a57d0c-c1b3-42e4-c028-8168f63097f7"
      },
      "source": [
        "# Cross-Validation com Balanceamento RUS e RobustScaler\n",
        "print('Cross-validation RF:')\n",
        "score_teste1 = val_model_balanced(X_rus_Rscaled, y_train_rus_Rscaled, rf)\n",
        "print('\\nCross-validation LR:')\n",
        "score_teste2 = val_model_balanced(X_rus_Rscaled, y_train_rus_Rscaled, lr)\n",
        "print('\\nCross-validation XGB:')\n",
        "score_teste3 = val_model_balanced(X_rus_Rscaled, y_train_rus_Rscaled, xgb)\n",
        "print('\\nCross-validation LDA:')\n",
        "score_teste4 = val_model_balanced(X_rus_Rscaled, y_train_rus_Rscaled, lda)\n",
        "print('\\nCross-validation SVC:')\n",
        "score_teste5 = val_model_balanced(X_rus_Rscaled, y_train_rus_Rscaled, svc)\n",
        "print('\\nCross-validation SGD:')\n",
        "score_teste6 = val_model_balanced(X_rus_Rscaled, y_train_rus_Rscaled, sgd)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Cross-validation RF:\n",
            "Recall: 0.7689 (+/- 0.0464)\n",
            "\n",
            "Cross-validation LR:\n",
            "Recall: 0.7981 (+/- 0.0338)\n",
            "\n",
            "Cross-validation XGB:\n",
            "Recall: 0.8010 (+/- 0.0529)\n",
            "\n",
            "Cross-validation LDA:\n",
            "Recall: 0.7974 (+/- 0.0275)\n",
            "\n",
            "Cross-validation SVC:\n",
            "Recall: 0.7867 (+/- 0.0307)\n",
            "\n",
            "Cross-validation SGD:\n",
            "Recall: 0.7561 (+/- 0.1035)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "aQ8AnR2VOzGU"
      },
      "source": [
        "### Balanceamento RUS com QuantileTransformer (QTR)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "kHWchUcZJnLv",
        "outputId": "0c080fe1-97da-4fef-ad5e-34b4c709a2e9"
      },
      "source": [
        "# Cross-Validation com Balanceamento RUS e Quantile Transformer\n",
        "print('Cross-validation RF:')\n",
        "score_teste1 = val_model_balanced(X_rus_qtr, y_train_rus_qtr, rf)\n",
        "print('\\nCross-validation LR:')\n",
        "score_teste2 = val_model_balanced(X_rus_qtr, y_train_rus_qtr, lr)\n",
        "print('\\nCross-validation XGB:')\n",
        "score_teste3 = val_model_balanced(X_rus_qtr, y_train_rus_qtr, xgb)\n",
        "print('\\nCross-validation LDA:')\n",
        "score_teste4 = val_model_balanced(X_rus_qtr, y_train_rus_qtr, lda)\n",
        "print('\\nCross-validation SVC:')\n",
        "score_teste5 = val_model_balanced(X_rus_qtr, y_train_rus_qtr, svc)\n",
        "print('\\nCross-validation SGD:')\n",
        "score_teste6 = val_model_balanced(X_rus_qtr, y_train_rus_qtr, sgd)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Cross-validation RF:\n",
            "Recall: 0.7746 (+/- 0.0412)\n",
            "\n",
            "Cross-validation LR:\n",
            "Recall: 0.7946 (+/- 0.0330)\n",
            "\n",
            "Cross-validation XGB:\n",
            "Recall: 0.8096 (+/- 0.0260)\n",
            "\n",
            "Cross-validation LDA:\n",
            "Recall: 0.8046 (+/- 0.0221)\n",
            "\n",
            "Cross-validation SVC:\n",
            "Recall: 0.8088 (+/- 0.0319)\n",
            "\n",
            "Cross-validation SGD:\n",
            "Recall: 0.8233 (+/- 0.3374)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "lT7PMN4MO_Yw"
      },
      "source": [
        "### Balanceamento RUS com QuantileTransformer (uniform output)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "S3GiB-d0LywL",
        "outputId": "f7cdc5bb-26d9-4991-f740-81c4e53e8d35"
      },
      "source": [
        "# Cross-Validation com Balanceamento RUS e Quantile Transformer (uniform output)\n",
        "print('Cross-validation RF:')\n",
        "score_teste1 = val_model_balanced(X_rus_qtr_n, y_train_rus_qtr_n, rf)\n",
        "print('\\nCross-validation LR:')\n",
        "score_teste2 = val_model_balanced(X_rus_qtr_n, y_train_rus_qtr_n, lr)\n",
        "print('\\nCross-validation XGB:')\n",
        "score_teste3 = val_model_balanced(X_rus_qtr_n, y_train_rus_qtr_n, xgb)\n",
        "print('\\nCross-validation LDA:')\n",
        "score_teste4 = val_model_balanced(X_rus_qtr_n, y_train_rus_qtr_n, lda)\n",
        "print('\\nCross-validation SVC:')\n",
        "score_teste5 = val_model_balanced(X_rus_qtr_n, y_train_rus_qtr_n, svc)\n",
        "print('\\nCross-validation SGD:')\n",
        "score_teste6 = val_model_balanced(X_rus_qtr_n, y_train_rus_qtr_n, sgd)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Cross-validation RF:\n",
            "Recall: 0.7667 (+/- 0.0340)\n",
            "\n",
            "Cross-validation LR:\n",
            "Recall: 0.8181 (+/- 0.0384)\n",
            "\n",
            "Cross-validation XGB:\n",
            "Recall: 0.8138 (+/- 0.0453)\n",
            "\n",
            "Cross-validation LDA:\n",
            "Recall: 0.8267 (+/- 0.0365)\n",
            "\n",
            "Cross-validation SVC:\n",
            "Recall: 0.8110 (+/- 0.0294)\n",
            "\n",
            "Cross-validation SGD:\n",
            "Recall: 0.6886 (+/- 0.4249)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "f4hRYw1wPpMM"
      },
      "source": [
        "### Balanceamento RUS com PowerTransformer (PWR)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "q7YbsKUALzOd",
        "outputId": "8be9e212-3e72-41a5-caf4-20e2fb92e0aa"
      },
      "source": [
        "# Cross-Validation com Balanceamento RUS e Power Transformer\n",
        "print('Cross-validation RF:')\n",
        "score_teste1 = val_model_balanced(X_rus_pwr, y_train_rus_pwr, rf)\n",
        "print('\\nCross-validation LR:')\n",
        "score_teste2 = val_model_balanced(X_rus_pwr, y_train_rus_pwr, lr)\n",
        "print('\\nCross-validation XGB:')\n",
        "score_teste3 = val_model_balanced(X_rus_pwr, y_train_rus_pwr, xgb)\n",
        "print('\\nCross-validation LDA:')\n",
        "score_teste4 = val_model_balanced(X_rus_pwr, y_train_rus_pwr, lda)\n",
        "print('\\nCross-validation SVC:')\n",
        "score_teste5 = val_model_balanced(X_rus_pwr, y_train_rus_pwr, svc)\n",
        "print('\\nCross-validation SGD:')\n",
        "score_teste6 = val_model_balanced(X_rus_pwr, y_train_rus_pwr, sgd)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Cross-validation RF:\n",
            "Recall: 0.7739 (+/- 0.0499)\n",
            "\n",
            "Cross-validation LR:\n",
            "Recall: 0.8038 (+/- 0.0325)\n",
            "\n",
            "Cross-validation XGB:\n",
            "Recall: 0.8352 (+/- 0.0444)\n",
            "\n",
            "Cross-validation LDA:\n",
            "Recall: 0.8096 (+/- 0.0280)\n",
            "\n",
            "Cross-validation SVC:\n",
            "Recall: 0.8067 (+/- 0.0364)\n",
            "\n",
            "Cross-validation SGD:\n",
            "Recall: 0.7676 (+/- 0.1522)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "1m2iNLCCP-zO"
      },
      "source": [
        "## Balanceamento SMOTE com StandardScaler (scaled)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "IkCjzFY6PGCv",
        "outputId": "1e8274bd-36e8-4d99-90c9-7cacdbee9187"
      },
      "source": [
        "# Cross-Validation com Balanceamento SMOTE e Standard Scaler \n",
        "print('Cross-validation RF:')\n",
        "score_teste1 = val_model_balanced(X_smo_scaled, y_train_smo_scaled, rf)\n",
        "print('\\nCross-validation LR:')\n",
        "score_teste2 = val_model_balanced(X_smo_scaled, y_train_smo_scaled, lr)\n",
        "print('\\nCross-validation XGB:')\n",
        "score_teste3 = val_model_balanced(X_smo_scaled, y_train_smo_scaled, xgb)\n",
        "print('\\nCross-validation LDA:')\n",
        "score_teste4 = val_model_balanced(X_smo_scaled, y_train_smo_scaled, lda)\n",
        "print('\\nCross-validation SVC:')\n",
        "score_teste5 = val_model_balanced(X_smo_scaled, y_train_smo_scaled, svc)\n",
        "print('\\nCross-validation SGD:')\n",
        "score_teste6 = val_model_balanced(X_smo_scaled, y_train_smo_scaled, sgd)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Cross-validation RF:\n",
            "Recall: 0.8719 (+/- 0.2336)\n",
            "\n",
            "Cross-validation LR:\n",
            "Recall: 0.8178 (+/- 0.0381)\n",
            "\n",
            "Cross-validation XGB:\n",
            "Recall: 0.8557 (+/- 0.2815)\n",
            "\n",
            "Cross-validation LDA:\n",
            "Recall: 0.8204 (+/- 0.0465)\n",
            "\n",
            "Cross-validation SVC:\n",
            "Recall: 0.8451 (+/- 0.1068)\n",
            "\n",
            "Cross-validation SGD:\n",
            "Recall: 0.8034 (+/- 0.1332)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Uv360uUPQIfv"
      },
      "source": [
        "#### Balanceamento SMOTE com RobustScaler (Rscaled)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "rWXW1my1QFqe",
        "outputId": "9f7bdb0f-372c-4193-bd2e-49e050680dda"
      },
      "source": [
        "# Cross-Validation com Balanceamento SMOTE e Robust Scaler\n",
        "print('Cross-validation RF:')\n",
        "score_teste1 = val_model_balanced(X_smo_Rscaled, y_train_smo_Rscaled, rf)\n",
        "print('\\nCross-validation LR:')\n",
        "score_teste2 = val_model_balanced(X_smo_Rscaled, y_train_smo_Rscaled, lr)\n",
        "print('\\nCross-validation XGB:')\n",
        "score_teste3 = val_model_balanced(X_smo_Rscaled, y_train_smo_Rscaled, xgb)\n",
        "print('\\nCross-validation LDA:')\n",
        "score_teste4 = val_model_balanced(X_smo_Rscaled, y_train_smo_Rscaled, lda)\n",
        "print('\\nCross-validation SVC:')\n",
        "score_teste5 = val_model_balanced(X_smo_Rscaled, y_train_smo_Rscaled, svc)\n",
        "print('\\nCross-validation SGD:')\n",
        "score_teste6 = val_model_balanced(X_smo_Rscaled, y_train_smo_Rscaled, sgd)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Cross-validation RF:\n",
            "Recall: 0.8768 (+/- 0.2312)\n",
            "\n",
            "Cross-validation LR:\n",
            "Recall: 0.8216 (+/- 0.0331)\n",
            "\n",
            "Cross-validation XGB:\n",
            "Recall: 0.8595 (+/- 0.2726)\n",
            "\n",
            "Cross-validation LDA:\n",
            "Recall: 0.8168 (+/- 0.0373)\n",
            "\n",
            "Cross-validation SVC:\n",
            "Recall: 0.8472 (+/- 0.1149)\n",
            "\n",
            "Cross-validation SGD:\n",
            "Recall: 0.8314 (+/- 0.0641)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "kY68UgTXQUGj"
      },
      "source": [
        "#### Balanceamento SMOTE com QuantileTransformer (QTR)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "WscKfsNyQNLh",
        "outputId": "c718e458-d8e3-415a-9a42-a9aaaad0c9fe"
      },
      "source": [
        "# Cross-Validation com Balanceamento SMOTE e Quantile Transformer\n",
        "print('Cross-validation RF:')\n",
        "score_teste1 = val_model_balanced(X_smo_qtr, y_train_smo_qtr, rf)\n",
        "print('\\nCross-validation LR:')\n",
        "score_teste2 = val_model_balanced(X_smo_qtr, y_train_smo_qtr, lr)\n",
        "print('\\nCross-validation XGB:')\n",
        "score_teste3 = val_model_balanced(X_smo_qtr, y_train_smo_qtr, xgb)\n",
        "print('\\nCross-validation LDA:')\n",
        "score_teste4 = val_model_balanced(X_smo_qtr, y_train_smo_qtr, lda)\n",
        "print('\\nCross-validation SVC:')\n",
        "score_teste5 = val_model_balanced(X_smo_qtr, y_train_smo_qtr, svc)\n",
        "print('\\nCross-validation SGD:')\n",
        "score_teste6 = val_model_balanced(X_smo_qtr, y_train_smo_qtr, sgd)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Cross-validation RF:\n",
            "Recall: 0.8691 (+/- 0.2326)\n",
            "\n",
            "Cross-validation LR:\n",
            "Recall: 0.8090 (+/- 0.0461)\n",
            "\n",
            "Cross-validation XGB:\n",
            "Recall: 0.8631 (+/- 0.2649)\n",
            "\n",
            "Cross-validation LDA:\n",
            "Recall: 0.8139 (+/- 0.0408)\n",
            "\n",
            "Cross-validation SVC:\n",
            "Recall: 0.8631 (+/- 0.1367)\n",
            "\n",
            "Cross-validation SGD:\n",
            "Recall: 0.7711 (+/- 0.1670)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "pS5_TN1kQaiQ"
      },
      "source": [
        "#### Balanceamento SMOTE com QuantileTransformer (uniform output)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "1htMBihgQokC",
        "outputId": "3d0fd5a8-6f9d-436a-e7e4-7b4865c0bbea"
      },
      "source": [
        "# Cross-Validation com Balanceamento SMOTE e Quantile Transformer (uniform Output)\n",
        "print('Cross-validation RF:')\n",
        "score_teste1 = val_model_balanced(X_smo_qtr_n, y_train_smo_qtr_n, rf)\n",
        "print('\\nCross-validation LR:')\n",
        "score_teste2 = val_model_balanced(X_smo_qtr_n, y_train_smo_qtr_n, lr)\n",
        "print('\\nCross-validation XGB:')\n",
        "score_teste3 = val_model_balanced(X_smo_qtr_n, y_train_smo_qtr_n, xgb)\n",
        "print('\\nCross-validation LDA:')\n",
        "score_teste4 = val_model_balanced(X_smo_qtr_n, y_train_smo_qtr_n, lda)\n",
        "print('\\nCross-validation SVC:')\n",
        "score_teste5 = val_model_balanced(X_smo_qtr_n, y_train_smo_qtr_n, svc)\n",
        "print('\\nCross-validation SGD:')\n",
        "score_teste6 = val_model_balanced(X_smo_qtr_n, y_train_smo_qtr_n, sgd)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Cross-validation RF:\n",
            "Recall: 0.8753 (+/- 0.2161)\n",
            "\n",
            "Cross-validation LR:\n",
            "Recall: 0.8165 (+/- 0.0275)\n",
            "\n",
            "Cross-validation XGB:\n",
            "Recall: 0.8662 (+/- 0.2547)\n",
            "\n",
            "Cross-validation LDA:\n",
            "Recall: 0.8322 (+/- 0.0290)\n",
            "\n",
            "Cross-validation SVC:\n",
            "Recall: 0.8492 (+/- 0.1279)\n",
            "\n",
            "Cross-validation SGD:\n",
            "Recall: 0.7013 (+/- 0.2926)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "dPaKsL2hQgWT"
      },
      "source": [
        "#### Balanceamento SMOTE com PowerTransformer (PWR)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "VaGZN7FuQ_LW",
        "outputId": "f4036268-f94b-4a7a-e5ae-12087bfa0de4"
      },
      "source": [
        "# Cross-Validation com Balanceamento SMOTE e Power Transformer\n",
        "print('Cross-validation RF:')\n",
        "score_teste1 = val_model_balanced(X_smo_pwr, y_train_smo_pwr, rf)\n",
        "print('\\nCross-validation LR:')\n",
        "score_teste2 = val_model_balanced(X_smo_pwr, y_train_smo_pwr, lr)\n",
        "print('\\nCross-validation XGB:')\n",
        "score_teste3 = val_model_balanced(X_smo_pwr, y_train_smo_pwr, xgb)\n",
        "print('\\nCross-validation LDA:')\n",
        "score_teste4 = val_model_balanced(X_smo_pwr, y_train_smo_pwr, lda)\n",
        "print('\\nCross-validation SVC:')\n",
        "score_teste5 = val_model_balanced(X_smo_pwr, y_train_smo_pwr, svc)\n",
        "print('\\nCross-validation SGD:')\n",
        "score_teste6 = val_model_balanced(X_smo_pwr, y_train_smo_pwr, sgd)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Cross-validation RF:\n",
            "Recall: 0.8719 (+/- 0.2344)\n",
            "\n",
            "Cross-validation LR:\n",
            "Recall: 0.8101 (+/- 0.0531)\n",
            "\n",
            "Cross-validation XGB:\n",
            "Recall: 0.8595 (+/- 0.2960)\n",
            "\n",
            "Cross-validation LDA:\n",
            "Recall: 0.8204 (+/- 0.0536)\n",
            "\n",
            "Cross-validation SVC:\n",
            "Recall: 0.8644 (+/- 0.1278)\n",
            "\n",
            "Cross-validation SGD:\n",
            "Recall: 0.7353 (+/- 0.1202)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "qqA-rXHZQmZx"
      },
      "source": [
        "## Balanceamento ADASYN com StandardScaler (scaled)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "dmQE1mTWRU0D",
        "outputId": "c87283f7-26d7-4c89-9599-47a6e5835199"
      },
      "source": [
        "# Cross-Validation com Balanceamento ADASYN e Standard Scaler\n",
        "print('Cross-validation RF:')\n",
        "score_teste1 = val_model_balanced(X_ada_scaled, y_train_ada_scaled, rf)\n",
        "print('\\nCross-validation LR:')\n",
        "score_teste2 = val_model_balanced(X_ada_scaled, y_train_ada_scaled, lr)\n",
        "print('\\nCross-validation XGB:')\n",
        "score_teste3 = val_model_balanced(X_ada_scaled, y_train_ada_scaled, xgb)\n",
        "print('\\nCross-validation LDA:')\n",
        "score_teste4 = val_model_balanced(X_ada_scaled, y_train_ada_scaled, lda)\n",
        "print('\\nCross-validation SVC:')\n",
        "score_teste5 = val_model_balanced(X_ada_scaled, y_train_ada_scaled, svc)\n",
        "print('\\nCross-validation SGD:')\n",
        "score_teste6 = val_model_balanced(X_ada_scaled, y_train_ada_scaled, sgd)\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Cross-validation RF:\n",
            "Recall: 0.8754 (+/- 0.1974)\n",
            "\n",
            "Cross-validation LR:\n",
            "Recall: 0.7970 (+/- 0.0914)\n",
            "\n",
            "Cross-validation XGB:\n",
            "Recall: 0.8536 (+/- 0.2742)\n",
            "\n",
            "Cross-validation LDA:\n",
            "Recall: 0.7849 (+/- 0.0986)\n",
            "\n",
            "Cross-validation SVC:\n",
            "Recall: 0.8703 (+/- 0.0440)\n",
            "\n",
            "Cross-validation SGD:\n",
            "Recall: 0.7898 (+/- 0.0628)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "tSTmGoGDQ4tD"
      },
      "source": [
        "#### Balanceamento ADASYN com RobustScaler (Rscaled)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "SdMrW6JXRw0q",
        "outputId": "4fed1cf8-a11c-457b-895b-f14bd419de99"
      },
      "source": [
        "# Cross-Validation com Balanceamento ADASYN e Robust Scaler\n",
        "print('Cross-validation RF:')\n",
        "score_teste1 = val_model_balanced(X_ada_Rscaled, y_train_ada_Rscaled, rf)\n",
        "print('\\nCross-validation LR:')\n",
        "score_teste2 = val_model_balanced(X_ada_Rscaled, y_train_ada_Rscaled, lr)\n",
        "print('\\nCross-validation XGB:')\n",
        "score_teste3 = val_model_balanced(X_ada_Rscaled, y_train_ada_Rscaled, xgb)\n",
        "print('\\nCross-validation LDA:')\n",
        "score_teste4 = val_model_balanced(X_ada_Rscaled, y_train_ada_Rscaled, lda)\n",
        "print('\\nCross-validation SVC:')\n",
        "score_teste5 = val_model_balanced(X_ada_Rscaled, y_train_ada_Rscaled, svc)\n",
        "print('\\nCross-validation SGD:')\n",
        "score_teste6 = val_model_balanced(X_ada_Rscaled, y_train_ada_Rscaled, sgd)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Cross-validation RF:\n",
            "Recall: 0.8795 (+/- 0.1876)\n",
            "\n",
            "Cross-validation LR:\n",
            "Recall: 0.7912 (+/- 0.0895)\n",
            "\n",
            "Cross-validation XGB:\n",
            "Recall: 0.8531 (+/- 0.2658)\n",
            "\n",
            "Cross-validation LDA:\n",
            "Recall: 0.7837 (+/- 0.0987)\n",
            "\n",
            "Cross-validation SVC:\n",
            "Recall: 0.8699 (+/- 0.0457)\n",
            "\n",
            "Cross-validation SGD:\n",
            "Recall: 0.7665 (+/- 0.1661)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "BKvImPxYRCWH"
      },
      "source": [
        "#### Balanceamento ADASYN com QuantileTransformer (QTR)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "nB3gr4wdSBQo",
        "outputId": "9f15e849-b285-4bd5-b4cd-6f556bb02a27"
      },
      "source": [
        "# Cross-Validation com Balanceamento ADASYN e Quantile Transformer \n",
        "print('Cross-validation RF:')\n",
        "score_teste1 = val_model_balanced(X_ada_qtr, y_train_ada_qtr, rf)\n",
        "print('\\nCross-validation LR:')\n",
        "score_teste2 = val_model_balanced(X_ada_qtr, y_train_ada_qtr, lr)\n",
        "print('\\nCross-validation XGB:')\n",
        "score_teste3 = val_model_balanced(X_ada_qtr, y_train_ada_qtr, xgb)\n",
        "print('\\nCross-validation LDA:')\n",
        "score_teste4 = val_model_balanced(X_ada_qtr, y_train_ada_qtr, lda)\n",
        "print('\\nCross-validation SVC:')\n",
        "score_teste5 = val_model_balanced(X_ada_qtr, y_train_ada_qtr, svc)\n",
        "print('\\nCross-validation SGD:')\n",
        "score_teste6 = val_model_balanced(X_ada_qtr, y_train_ada_qtr, sgd)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Cross-validation RF:\n",
            "Recall: 0.8839 (+/- 0.1847)\n",
            "\n",
            "Cross-validation LR:\n",
            "Recall: 0.7772 (+/- 0.0690)\n",
            "\n",
            "Cross-validation XGB:\n",
            "Recall: 0.8566 (+/- 0.2608)\n",
            "\n",
            "Cross-validation LDA:\n",
            "Recall: 0.7793 (+/- 0.0714)\n",
            "\n",
            "Cross-validation SVC:\n",
            "Recall: 0.8615 (+/- 0.0845)\n",
            "\n",
            "Cross-validation SGD:\n",
            "Recall: 0.7373 (+/- 0.2355)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "WhsA3tVCRH1Y"
      },
      "source": [
        "#### Balanceamento ADASYN com QuantileTransformer (uniform output)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "WVyRWGwTSXs6",
        "outputId": "0b111508-f558-4f37-f2d7-fce5f892b499"
      },
      "source": [
        "# Cross-Validation com Balanceamento ADASYN e Quantile Transformer (uniform output)\n",
        "print('Cross-validation RF:')\n",
        "score_teste1 = val_model_balanced(X_ada_qtr_n, y_train_ada_qtr_n, rf)\n",
        "print('\\nCross-validation LR:')\n",
        "score_teste2 = val_model_balanced(X_ada_qtr_n, y_train_ada_qtr_n, lr)\n",
        "print('\\nCross-validation XGB:')\n",
        "score_teste3 = val_model_balanced(X_ada_qtr_n, y_train_ada_qtr_n, xgb)\n",
        "print('\\nCross-validation LDA:')\n",
        "score_teste4 = val_model_balanced(X_ada_qtr_n, y_train_ada_qtr_n, lda)\n",
        "print('\\nCross-validation SVC:')\n",
        "score_teste5 = val_model_balanced(X_ada_qtr_n, y_train_ada_qtr_n, svc)\n",
        "print('\\nCross-validation SGD:')\n",
        "score_teste6 = val_model_balanced(X_ada_qtr_n, y_train_ada_qtr_n, sgd)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Cross-validation RF:\n",
            "Recall: 0.8749 (+/- 0.2001)\n",
            "\n",
            "Cross-validation LR:\n",
            "Recall: 0.7929 (+/- 0.0707)\n",
            "\n",
            "Cross-validation XGB:\n",
            "Recall: 0.8580 (+/- 0.2667)\n",
            "\n",
            "Cross-validation LDA:\n",
            "Recall: 0.7962 (+/- 0.0756)\n",
            "\n",
            "Cross-validation SVC:\n",
            "Recall: 0.8695 (+/- 0.0978)\n",
            "\n",
            "Cross-validation SGD:\n",
            "Recall: 0.6560 (+/- 0.3559)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "THnD77BARQaL"
      },
      "source": [
        "#### Balanceamento ADASYN com PowerTransformer (PWR)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "lLJYkvOaSfB6",
        "outputId": "929287c2-4562-4a69-97b5-c60bf82fee84"
      },
      "source": [
        "# Cross-Validation com Balanceamento ADASYN e Power Transformer\n",
        "print('Cross-validation RF:')\n",
        "score_teste1 = val_model_balanced(X_ada_pwr, y_train_ada_pwr, rf)\n",
        "print('\\nCross-validation LR:')\n",
        "score_teste2 = val_model_balanced(X_ada_pwr, y_train_ada_pwr, lr)\n",
        "print('\\nCross-validation XGB:')\n",
        "score_teste3 = val_model_balanced(X_ada_pwr, y_train_ada_pwr, xgb)\n",
        "print('\\nCross-validation LDA:')\n",
        "score_teste4 = val_model_balanced(X_ada_pwr, y_train_ada_pwr, lda)\n",
        "print('\\nCross-validation SVC:')\n",
        "score_teste5 = val_model_balanced(X_ada_pwr, y_train_ada_pwr, svc)\n",
        "print('\\nCross-validation SGD:')\n",
        "score_teste6 = val_model_balanced(X_ada_pwr, y_train_ada_pwr, sgd)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Cross-validation RF:\n",
            "Recall: 0.8715 (+/- 0.2031)\n",
            "\n",
            "Cross-validation LR:\n",
            "Recall: 0.7685 (+/- 0.0933)\n",
            "\n",
            "Cross-validation XGB:\n",
            "Recall: 0.8536 (+/- 0.2899)\n",
            "\n",
            "Cross-validation LDA:\n",
            "Recall: 0.7729 (+/- 0.0931)\n",
            "\n",
            "Cross-validation SVC:\n",
            "Recall: 0.8735 (+/- 0.0399)\n",
            "\n",
            "Cross-validation SGD:\n",
            "Recall: 0.7690 (+/- 0.1376)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "9KdaXNhaRZAl"
      },
      "source": [
        "## Definindo Modelos para o Balanceamento RUS\n",
        "\n",
        "Após realizar a validação de todos os modelos, com os 3 tipos de balanceamento e os 5 tipos de padronização, selecionei alguns modelos que obtiveram o melhor desempenho para o Balanceamento RUS. \n",
        "\n",
        "Serão usados os modelos:\n",
        "\n",
        "\n",
        "* Linear Discriminant Analysis (LDA) - com padronização RobustScaler\n",
        "\n",
        "* SVC - com padronização RobustScaler\n",
        "\n",
        "* XGBoost (XBG) - com padronização RobustScaler   \n",
        "\n",
        "Todos os três modelos passarão por um processo de otimização de seus respectivos hiperparâmetros, a fim de obtermos resultados ainda mais consistentes.\n",
        "\n",
        "\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 514
        },
        "id": "BcUV9n5pMWSr",
        "outputId": "570ceaf9-58b9-499c-baae-4d10a5882bbf"
      },
      "source": [
        "# Imprimir DataSet com os Dados de Recall dos Modelos Escolhidos\n",
        "model = []\n",
        "recall= []\n",
        "\n",
        "model.append('LinearDiscriminantAnalysis RusScaled')\n",
        "recall.append(val_model_balanced(X_rus_scaled, y_train_rus_scaled, lda, quite=True))  \n",
        "model.append('LinearDiscriminantAnalysis RusRscaled')\n",
        "recall.append(val_model_balanced(X_rus_Rscaled, y_train_rus_Rscaled, lda, quite=True))\n",
        "model.append('LinearDiscriminantAnalysis RusQTR')\n",
        "recall.append(val_model_balanced(X_rus_qtr, y_train_rus_qtr, lda, quite=True))\n",
        "model.append('LinearDiscriminantAnalysis RusQTR-N')\n",
        "recall.append(val_model_balanced(X_rus_qtr_n, y_train_rus_qtr_n, lda, quite=True))\n",
        "model.append('LinearDiscriminantAnalysis RusPWR')\n",
        "recall.append(val_model_balanced(X_rus_pwr, y_train_rus_pwr, lda, quite=True))\n",
        "\n",
        "model.append('SVC RusScaled')\n",
        "recall.append(val_model_balanced(X_rus_scaled, y_train_rus_scaled, svc, quite=True))\n",
        "model.append('SVC RusRscaled')\n",
        "recall.append(val_model_balanced(X_rus_Rscaled, y_train_rus_Rscaled, svc, quite=True))\n",
        "model.append('SVC RusQTR')\n",
        "recall.append(val_model_balanced(X_rus_qtr, y_train_rus_qtr, svc, quite=True))\n",
        "model.append('SVC RusQTR-N')\n",
        "recall.append(val_model_balanced(X_rus_qtr_n, y_train_rus_qtr_n, svc, quite=True))\n",
        "model.append('SVC RusPWR')\n",
        "recall.append(val_model_balanced(X_rus_pwr, y_train_rus_pwr, svc, quite=True))\n",
        "\n",
        "model.append('XGBoost RusScaled')\n",
        "recall.append(val_model_balanced(X_rus_scaled, y_train_rus_scaled, xgb, quite=True))  \n",
        "model.append('XGBoost RusRscaled')\n",
        "recall.append(val_model_balanced(X_rus_Rscaled, y_train_rus_Rscaled, xgb, quite=True))\n",
        "model.append('XGBoost RusQTR')\n",
        "recall.append(val_model_balanced(X_rus_qtr, y_train_rus_qtr, xgb, quite=True))\n",
        "model.append('XGBoost RusQTR-N')\n",
        "recall.append(val_model_balanced(X_rus_qtr_n, y_train_rus_qtr_n, xgb, quite=True))\n",
        "model.append('XGBoost RusPWR')\n",
        "recall.append(val_model_balanced(X_rus_pwr, y_train_rus_pwr, xgb, quite=True))\n",
        "\n",
        "\n",
        "pd.DataFrame(data=recall, index=model, columns=['Recall'])"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>Recall</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>LinearDiscriminantAnalysis RusScaled</th>\n",
              "      <td>0.808846</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>LinearDiscriminantAnalysis RusRscaled</th>\n",
              "      <td>0.797433</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>LinearDiscriminantAnalysis RusQTR</th>\n",
              "      <td>0.804558</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>LinearDiscriminantAnalysis RusQTR-N</th>\n",
              "      <td>0.826680</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>LinearDiscriminantAnalysis RusPWR</th>\n",
              "      <td>0.809558</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>SVC RusScaled</th>\n",
              "      <td>0.800292</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>SVC RusRscaled</th>\n",
              "      <td>0.786734</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>SVC RusQTR</th>\n",
              "      <td>0.808849</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>SVC RusQTR-N</th>\n",
              "      <td>0.810991</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>SVC RusPWR</th>\n",
              "      <td>0.806711</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>XGBoost RusScaled</th>\n",
              "      <td>0.826678</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>XGBoost RusRscaled</th>\n",
              "      <td>0.800991</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>XGBoost RusQTR</th>\n",
              "      <td>0.809550</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>XGBoost RusQTR-N</th>\n",
              "      <td>0.813843</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>XGBoost RusPWR</th>\n",
              "      <td>0.835241</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "                                         Recall\n",
              "LinearDiscriminantAnalysis RusScaled   0.808846\n",
              "LinearDiscriminantAnalysis RusRscaled  0.797433\n",
              "LinearDiscriminantAnalysis RusQTR      0.804558\n",
              "LinearDiscriminantAnalysis RusQTR-N    0.826680\n",
              "LinearDiscriminantAnalysis RusPWR      0.809558\n",
              "SVC RusScaled                          0.800292\n",
              "SVC RusRscaled                         0.786734\n",
              "SVC RusQTR                             0.808849\n",
              "SVC RusQTR-N                           0.810991\n",
              "SVC RusPWR                             0.806711\n",
              "XGBoost RusScaled                      0.826678\n",
              "XGBoost RusRscaled                     0.800991\n",
              "XGBoost RusQTR                         0.809550\n",
              "XGBoost RusQTR-N                       0.813843\n",
              "XGBoost RusPWR                         0.835241"
            ]
          },
          "metadata": {},
          "execution_count": 39
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Wfj3kP8ESC8k"
      },
      "source": [
        "### Otimizando Hiperparâmetros Linear Discriminant Analysis (LDA)\n",
        "\n",
        "O modelo LDA possui poucos hiperparâmetros para serem otimizados, mas garanto que eles fazem diferença e nos darão um resultado bem interessante.\n",
        "\n",
        "Pode-se notar que:\n",
        "\n",
        "* O melhor desempenho do modelo foi com `Shrinkage: auto` e `solver: lsqr` \n",
        "* Com essas configurações obtemos um recall de 0.80\n",
        "\n",
        "É normal que o recall seja alto na hora de realizar a otimização dos hiperparâmetros, portanto não estranhe caso o recall seja menor na avaliação de desempenho do modelo."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "0x9aG61IAdRX",
        "outputId": "8530c81d-1d15-4618-cd0b-396233d65843"
      },
      "source": [
        "# Modelo a ser otimizado\n",
        "lda_rus_Rscaled = LinearDiscriminantAnalysis()\n",
        "\n",
        "# Parâmetros a serem otimizados\n",
        "param_grid = {\n",
        "    'solver':('svd', 'lsqr', 'eigen'),\n",
        "    'shrinkage':('auto', 'None')\n",
        "}\n",
        "\n",
        "# Identificar melhores parâmetros\n",
        "kfold = StratifiedKFold(n_splits=10, shuffle=True)\n",
        "grid_search = GridSearchCV(lda_rus_Rscaled, param_grid, scoring='recall', n_jobs=-1, cv=kfold)\n",
        "grid_result = grid_search.fit(X_rus_Rscaled, y_train_rus_Rscaled)\n",
        "\n",
        "# Ver resultados\n",
        "print(f'Melhor: {grid_result.best_score_} para {grid_result.best_params_}')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Melhor: 0.8052887537993921 para {'shrinkage': 'auto', 'solver': 'lsqr'}\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "_bOyaPetSZkp"
      },
      "source": [
        "#### Avaliando o Desempenho do Modelo LDA\n",
        "\n",
        "Vamos realizar o treino do modelo, após isso será realizada a avaliação do desempenho final do modelo com os dados de teste, assim é possível ter uma noção de como esse modelo lida e irá lidar com dados novos."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 486
        },
        "id": "UHMvs7wzDiQo",
        "outputId": "d6f277e9-1b58-4e6e-f16d-34bcbeed2166"
      },
      "source": [
        "# Modelo final\n",
        "lda_rus_Rscaled = LinearDiscriminantAnalysis(solver='lsqr', shrinkage='auto')\n",
        "\n",
        "# Treinar modelo\n",
        "lda_rus_Rscaled.fit(X_rus_Rscaled, y_train_rus_Rscaled)\n",
        "\n",
        "# Padronizar dados de Teste\n",
        "X_test_Rscaled = Rob_scaler.fit_transform(X_test)\n",
        "\n",
        "# Fazer previsões em cima dos dados de teste\n",
        "y_pred_lda = lda_rus_Rscaled.predict(X_test_Rscaled)\n",
        "\n",
        "# Plotar matriz de confusão\n",
        "skplt.metrics.plot_confusion_matrix(y_test, y_pred_lda)\n",
        "\n",
        "# Imprimir Relatório de Classificação\n",
        "print(classification_report(y_test, y_pred_lda, digits=4))\n",
        "\n",
        "# Imprimir área sob curva\n",
        "print('AUC: {:.4f}\\n'.format(roc_auc_score(y_test, y_pred_lda)))\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "              precision    recall  f1-score   support\n",
            "\n",
            "           0     0.9036    0.7102    0.7953      1294\n",
            "           1     0.4960    0.7901    0.6094       467\n",
            "\n",
            "    accuracy                         0.7314      1761\n",
            "   macro avg     0.6998    0.7502    0.7024      1761\n",
            "weighted avg     0.7955    0.7314    0.7460      1761\n",
            "\n",
            "AUC: 0.7502\n",
            "\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 2 Axes>"
            ]
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "CYdbs74bSh_U"
      },
      "source": [
        "### Otimizando Hiperparâmetros SVC\n",
        "\n",
        "O modelo SVC possui diversos parâmetros, uma quantidade considerável a mais que o modelo LDA.\n",
        "\n",
        "Devido a essa quantidade maior de parâmetros, podemos otimizar ainda mais o modelo e obter um resultado ainda melhor. Entretanto, esse modelo se sai melhor apenas na otimização.\n",
        "\n",
        "Neste primeiro momento, vou buscar quais seriam os melhores valores para os parâmetros `C` e `kernel`.\n",
        "\n",
        "* O parâmetro `Kernel` seleciona o tipo de hiperplano usado para separar os dados. O uso de 'linear' utilizará um hiperplano linear (uma linha no caso de dados 2D). O 'rbf' e o 'poly' usam um hiperplano não linear \n",
        "\n",
        "* `C` é o parâmetro de penalidade do termo de erro. Ele controla o trade off entre o limite de decisão suave e a classificação correta dos pontos de treinamento. \n",
        "\n",
        "\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "DUCCPAJdFs9o",
        "outputId": "fa97b666-f4f8-43ff-960c-1500ca2dd5a2"
      },
      "source": [
        "# Modelo a ser otimizado\n",
        "svc_rus_Rscaled = SVC()\n",
        "\n",
        "# Parâmetros a serem otimizados\n",
        "param_grid = {\n",
        "    'C':[0.01, 0.1, 1.0, 10],\n",
        "    'kernel':('linear', 'poly', 'rbf', 'sigmoid')\n",
        "      \n",
        "}\n",
        "\n",
        "# Identificar melhores parâmetros\n",
        "kfold = StratifiedKFold(n_splits=10, shuffle=True)\n",
        "grid_search = GridSearchCV(svc_rus_Rscaled, param_grid, scoring='recall', n_jobs=-1, cv=kfold)\n",
        "grid_result = grid_search.fit(X_rus_Rscaled, y_train_rus_Rscaled)\n",
        "\n",
        "# Ver resultados\n",
        "print(f'Melhor: {grid_result.best_score_} para {grid_result.best_params_}')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Melhor: 0.9236372847011145 para {'C': 0.01, 'kernel': 'poly'}\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "DIuiWmYAWIjT"
      },
      "source": [
        "Uma vez com os parâmetros `C` e `kernel` definidos e tendo um hiperplano não linear, podemos começar as buscas pelo melhor parâmetero `gamma` e aproveitar para buscar pelo parâmetro `degree`.\n",
        "\n",
        "* `gamma` é um parâmetro para hiperplanos não lineares. Quanto mais alto o valor `gamma`, ele tenta se ajustar exatamente ao conjunto de dados de treinamento, isso pode gerar um problema quando os valores são muito altos, causando o overfitting.\n",
        "\n",
        "* `degree` é um parâmetro usado quando o kernel é definido como 'poly'. É basicamente o grau do polinômio usado para encontrar o hiperplano para dividir os dados."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "DouY9HxiGjSr",
        "outputId": "b39a3097-8c10-429f-83bd-51b7930d06af"
      },
      "source": [
        "# Modelo a ser otimizado\n",
        "svc_rus_Rscaled = SVC(C=0.01, kernel='poly')\n",
        "\n",
        "# Parâmetros a serem otimizados\n",
        "param_grid = {\n",
        "    'degree':[3, 5, 7],\n",
        "    'gamma':[1, 1.5]\n",
        "      \n",
        "}\n",
        "\n",
        "# Identificar melhores parâmetros\n",
        "kfold = StratifiedKFold(n_splits=10, shuffle=True)\n",
        "grid_search = GridSearchCV(svc_rus_Rscaled, param_grid, scoring='recall', n_jobs=-1, cv=kfold)\n",
        "grid_result = grid_search.fit(X_rus_Rscaled, y_train_rus_Rscaled)\n",
        "\n",
        "# Ver resultados\n",
        "print(f'Melhor: {grid_result.best_score_} para {grid_result.best_params_}')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Melhor: 0.7859979736575481 para {'degree': 3, 'gamma': 1}\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "wMMWOb88Su_k"
      },
      "source": [
        "#### Avaliando o Desempenho do Modelo SVC\n",
        "\n",
        "É possível notar que houve uma diminuição no desempenho, mesmo com os parâmetros otimizados, mas com isso podemos ter um resultado mais consistente do que se os parâmetros fossem mantidos em seus valores padrão.\n",
        "\n",
        "Vamos treinar o modelo em cima do dataset de treino e testar seu desempenho com os dados de teste, vamos levar em consideração a métrica `recall`, assim pode-se saber se o modelo está se saindo bem em identificar o Churn."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 482
        },
        "id": "kQ7lAM_wHSJk",
        "outputId": "a78cd512-0725-4d00-8d3f-807844c1a15d"
      },
      "source": [
        "# Modelo final\n",
        "svc_rus_Rscaled = SVC(C=0.01, kernel='poly', degree=3, gamma=1)\n",
        "\n",
        "# Treinar Modelo\n",
        "svc_rus_Rscaled.fit(X_rus_Rscaled, y_train_rus_Rscaled)\n",
        "\n",
        "# Fazer previsões em cima dos dados de teste\n",
        "y_pred_SvcRus = svc_rus_Rscaled.predict(X_test_Rscaled)\n",
        "\n",
        "# Plotar matriz de confusão\n",
        "skplt.metrics.plot_confusion_matrix(y_test, y_pred_SvcRus)\n",
        "\n",
        "# Imprimir Relatório de Classificação\n",
        "print(classification_report(y_test, y_pred_SvcRus, digits=4))\n",
        "\n",
        "# Imprimir área sob curva\n",
        "print('AUC: {:.4f}\\n'.format(roc_auc_score(y_test, y_pred_SvcRus)))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "              precision    recall  f1-score   support\n",
            "\n",
            "           0     0.8929    0.6638    0.7615      1294\n",
            "           1     0.4556    0.7794    0.5750       467\n",
            "\n",
            "    accuracy                         0.6945      1761\n",
            "   macro avg     0.6743    0.7216    0.6683      1761\n",
            "weighted avg     0.7769    0.6945    0.7121      1761\n",
            "\n",
            "AUC: 0.7216\n",
            "\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAATgAAAEWCAYAAADy2YssAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAgAElEQVR4nO3de1yUZf7/8dfNUVBQhmDQJMtT+iW/ampKq5IYooICHr7Uryh1XbNcyUOaZeYRc81WV602skjNstU4eCoPWFAeEFfRFrVW00KFYeUgiMphuH9/sM7GKsyMgMMMn2ePeeTc9z339Rl88Pa67tOlqKqqIoQQNsjO0gUIIURDkYATQtgsCTghhM2SgBNC2CwJOCGEzZKAE0LYLAk4G3Pz5k0mT55Mr169iI6Ovuv9bNu2jQkTJtRjZZYxceJEEhISLF2GsBBFroOzjO3btxMXF8f58+dp3rw5Xbp0YfLkyfTu3btO+01MTOTTTz9l8+bNODg41FO19SctLY3nnnuOJ598knfffdew/MyZM4SFhfHYY4+xceNGo/tZs2YNv/zyCytWrGjIcoWVa3y/AU1AXFwcsbGxLFy4kP79++Po6Mh3331HcnJynQPu8uXLPPjgg40y3G7RaDRkZGRQUFCAh4cHAAkJCTz44IP11oaqqqiqip2dDFKaMvnbv8eKi4tZvXo1b775JkOGDMHV1RVHR0cCAwN59dVXASgrKyMmJob+/fvTv39/YmJiKCsrA6p6QAMHDuTjjz/G39+f/v378+WXXwKwevVq3nvvPb766it69uzJli1bWLNmDa+88oqh/YsXL/Lwww9TUVEBQHx8PIMHD6Znz54EBgaybds2w/Knn37a8Lljx44xevRoevXqxejRozl27JhhXVRUFKtWreKpp56iZ8+eTJgwgfz8/Bp/Bo6OjgwePJhdu3YBoNfr2bVrFyNGjKi23ZIlSwgICODRRx9l1KhRHD16FIDU1FQ++OADw/ccOXKkoY6VK1fy1FNP0b17d7KysoiKimLLli0AzJ8/n6lTpxr2//bbb/P8888jgxjbJQF3jx0/fpzS0lKCgoJq3Ob999/nxIkTJCUlsW3bNn744Qfee+89w/orV65QXFxMamoqMTExLFq0iKtXrxIdHc0LL7zAsGHDOH78OGPHjq21luvXr7NkyRI+/PBDjh8/zubNm+natett2xUWFvLCCy8QFRVFWloa48eP54UXXqCgoMCwzY4dO3jrrbc4dOgQ5eXlfPzxx7W2HR4eTmJiIgDff/89nTt3RqvVVtumW7duJCYmcuTIEUJDQ3n55ZcpLS1l4MCB1b7nrVAGSEpKYvHixRw7dow2bdpU29+cOXP46aefiI+P5+jRo2zdupU//elPKIpSa63CeknA3WOFhYV4eHjUOoTcvn07U6ZMwdPTE41Gw5QpU6r9Ejs4ODBlyhQcHR0JCAjA1dWV8+fP31U9dnZ2/POf/+TmzZt4e3vTqVOn27b59ttvadeuHeHh4Tg4OBAaGkr79u355ptvDNuMGjWKhx56iGbNmjF06FBOnz5da7uPPvooV69e5eeffyYxMZGwsLDbtgkLCzP8rCZMmEBZWZnR7xkREUGnTp1wcHDA0dGx2joXFxeWL1/OsmXLmDVrFvPmzcPHx6fW/QnrJgF3j7Vq1YqCggLDEPFOcnNzq/U+2rRpQ25ubrV9/DYgXVxcuH79utm1uLq6snLlSjZv3kz//v2ZNGkS586dM1rPrZp0Op3hvZeXl9n1jBw5kk2bNpGWlnbHHu1HH33EsGHD6NWrF71796a4uLhar/FOWrduXev67t2707ZtW1RVZdiwYUZrFNZNAu4e69mzJ05OTuzbt6/Gbby9vbl8+bLhfXZ2Nt7e3nfVnouLCzdv3jS8v3LlSrX1AwYMIC4uju+//5727dszb948o/Xcqum/h5TmCgsL47PPPiMgIAAXF5dq644ePcq6detYtWoV6enpHD16FDc3N8PxspqGlcaGm5s2baK8vBxvb2/WrVtXp/pF4ycBd4+5ubkRHR3NokWL2LdvHzdu3KC8vJyUlBSWL18OQEhICO+//z75+fnk5+fz7rvv3nYA3lRdu3YlPT2dy5cvU1xczAcffGBYd+XKFfbt28f169dxcnLC1dX1jmcdAwICuHDhAtu3b6eiooJdu3Zx9uxZnnjiibuq6RZfX182btzItGnTbltXUlKCvb09Go2GiooK1q5dy7Vr1wzrPT09uXTpEpWVlSa3d/78eVatWsXbb7/N8uXLWbdundGhtLBuEnAWMGHCBObMmcN7772Hv78/TzzxBJs2beLJJ58E4KWXXuKRRx5h5MiRjBw5Ej8/P1566aW7aut3v/sdw4cPZ+TIkYwaNYpBgwYZ1lVWVvLJJ58wYMAAHnvsMdLT01mwYMFt+/Dw8OCvf/0rcXFx9O3bl3Xr1vHXv/4VjUZzVzX9Vu/eve/YE+zfvz8DBgwgODiYwMBAnJ2dqw0/hw4dCkDfvn2JiIgw2k5FRQWzZs3iD3/4A126dOHBBx9k+vTpzJ4923CGWtgeudBXCGGzpAcnhLBZEnBCCJslASeEsFkScEIIm9Wo7si+UnCNX7JrvodRND6t297d9XnCMlo1c8DVqW6/9nsOnMLTo4VJ2/b6nwfq1FZdNaqA+yU7n/7PLLd0GcIMc9++/Ro20XhNeKxtnQPO06OFyb+nN46vrVNbddWoAk4IYSUU6zi6JQEnhDCTAnb2li7CJBJwQgjzWckjpiTghBDmUZAhqhDChkkPTghhmxTpwQkhbJj04IQQNkmRs6hCCFsmQ1QhhM2SIaoQwjbJSQYhhC2TgBNC2CRFAfv6OcnwySefsGXLFhRFoXPnzrz11lvk5uYyY8YMCgsL8fPzY/ny5Tg5OVFWVsbs2bPJzMykVatWrFy5krZt29a6f+uIYSFE46Iopr1qodPp2LBhA19++SU7duxAr9ezc+dOVqxYwbhx49i7dy/u7u5s3boVgC1btuDu7s7evXsZN24cK1asMFqmBJwQwnyKnWkvI/R6PTdv3qSiooKbN2/i5eXF4cOHCQ4OBiAiIoLk5GQA9u/fb5hBLTg4mEOHDmFsziwZogohzKNg8lnU/Px8Jk6caHgfGRlJZGQkAFqtlgkTJjBo0CCcnZ353e9+h5+fH+7u7jg4VEWTj48POp0OqOrx3Zo60sHBATc3NwoKCmqdvlICTghhJtPPomo0GuLj4++47urVqyQnJ5OcnIybmxsvv/wy3333XX0WKkNUIcRdqIdjcAcPHqRt27ZoNBocHR0ZMmQIx44do6ioiIqKCgBycnIME4NrtVqys7OBqom8i4uL8fDwqLUNCTghhJn+fauWKa9atGnThhMnTnDjxg1UVeXQoUN07NiRvn37snv3bgASEhIIDAwEIDAwkISEBAB2795Nv379UIyEqAScEMI8t54HV8eTDN27dyc4OJiIiAhGjBhBZWUlkZGRzJo1i7i4OIKCgigsLGTs2LEAjBkzhsLCQoKCgoiLi+OVV14xWqocgxNCmK+ebtWKjo4mOjq62jJfX1/DpSG/5ezszOrVq83avwScEMJMcquWEMKWScAJIWySPA9OCGHT5HFJQgibJUNUIYTNkh6cEMIWKShGL7BtLCTghBDmUZCAE0LYLsVOAk4IYaOkByeEsFkScEIIm6QocpJBCGHLrCPfJOCEEOaTHpwQwmbZ2cmdDEIIG6TIdXBCCJtmHfkmASeEMJecRRVC2DAJOCGEbVLkVi0hhA2rjx7czz//zPTp0w3vs7KyiI6OJjw8nOnTp3Pp0iXuv/9+Vq1aRcuWLVFVlZiYGFJSUmjWrBnLli3Dz8+v1jas41yvEKLRUPjP3QzGXrVp3749SUlJJCUlER8fj4uLC0FBQcTGxuLv78+ePXvw9/cnNjYWgNTUVC5cuMCePXtYvHgxCxYsMFqrBJwQwkymhZs5vbxDhw7h6+vL/fffT3JyMuHh4QCEh4ezb98+AMNyRVHo0aMHRUVF5Obm1rpfGaIKIcxjxnVw+fn5TJw40fA+MjKSyMjI27bbuXMnoaGhAOTl5eHt7Q2Al5cXeXl5AOh0Onx8fAyf8fHxQafTGba9Ewk4IYT5TOycaTQa4uPja92mrKyM/fv3M3PmzNubqeON/TJEFUKYzc7OzqSXKVJTU/Hz8+O+++4DwNPT0zD0zM3NRaPRAKDVasnJyTF8LicnB61WW3udd/PlhBBNl1LPx+B27txJSEiI4X1gYCCJiYkAJCYmMnjw4GrLVVUlIyMDNze3WoenIAEnhDCXYsbLiOvXr3Pw4EGGDBliWDZp0iQOHDjAkCFDOHjwIJMmTQIgICAAX19fgoKCmDdvHvPnzze6fzkGVw+mPjOIcRGPo6oqmWcvM2n+p6yZ+xQDenXk6rWbAEx6cyMnf7pEKzcXPljwLA+1vY/SsnJeWLCJU+eyLfwNmiYF+H1fX4pvVvDFiWxCu3rT2t0ZgPzr5Ww7paNcr/K/rd0Y3Ok+iksrADiadZWMy0UWrNzy6utOBldXV9LS0qot8/DwYP369Xds05RQ+60GDbjU1FRiYmKorKxk7NixhiS2JW28WvLS0wH0HB3DzdJyPv3TBMYG9wLg9VWJJOzLqLb97N8Hc+LHi0TO/JDOD2pZNef/GD55jSVKb/Iee6AVV0rKcLavGsjs+elflOlVAJ7sdB992rbk4C+FAJzSFbP7xysWq7WxsZZbtRpsiKrX61m0aBHr1q1j586d7Nixg7NnzzZUcxblYG+Pi7Mj9vZ2uDRzIvtfV2vctkt7H1LSfwLgpws62rXR4K1xu1elin9zc7an432uZFz6T0/sVrgBONorqHf6oADq50Lfe6HBAu7kyZO0a9cOX19fnJycCAkJITk5uaGas5jL/7rKqg3J/PTVYs7vjaHo2g2SD58BYMGUERz54jWWzxyFk2NVZ/mHny4RFtgdgN5+7XigtYb7ta0sVn9TNaSzF8n/zLstxEb8jzfTBjyIp6sj6Vn/+Yeqi3cL/tDXl9HdfHB3buJHdv59L6opL0trsID774vytFotOp2uoZqzmFZuLoQ+0Y2uofNpP2QuzV2ceGp4H95cs43uEYvp/+zbeLRszszxTwKwIm4vLd1cObx5Di8+FcCJHy+i11da+Fs0LR3vc6WkTE9Ocelt67afyuUv313gSkk5ftoWAPzzSglrv7/Ah2lZnM+/zki/2s/c2br6PovakJr4P0V1F9i3Cxcu53Gl4BoAiftP0K/7Q2zelQ5AWXkFG5IOM+25qlPdxSU3eWHBp4bPn9m5kPOX8u594U2Yb0sXOns1p+N9rjjYKTg72BHmpyUps+ofYJWqY27+7Tw4kV3MjfL//AN0/FIRgZ08LVR549EYwssUDRZw/31Rnk6nM3pRnjXKysnnsW4P4dLMkRs3yxn02MMcO/UrPve5k3Ol6vjOyEH/y6lzlwFo2cKF6zfLKK/QMz7icb4/dpbikpuW/ApNzjfn8vjmXNU/Ku08XOj3QCuSMnV4uDhScKMcgE5ezblyvQyAFk72XCvTA9DZqzlXSsotU3gjYiX51nAB161bNy5cuEBWVhZarZadO3fyzjvvNFRzFpP+j19I2HecQ5+9SoW+khNnLvLRlwdIWvsi93m4oShw8seLTI3ZDFSdZPhwURSqqnL6XDaTF26y8DcQt4z088bZoeqoTW5xGbvOVF1N38e3FZ29XKlU4Ua5nu2ZtneoxVzW0oNTVFVtsJNFKSkpLF26FL1ez+jRo3nxxRdr3f7vp36l/zPLG6oc0QDmvj3N0iUIM0x4rC1t3JvVaR//uHiV0WsOm7Ttj38KrlNbddWgx+ACAgIICAhoyCaEEBZgJR04OckghDCfXSO4BMQUEnBCCPMo0oMTQtgoRaYNFELYMivJNwk4IYT5TH2YpaVJwAkhzCPH4IQQturWtIHWQAJOCGE2K8k3CTghhPmkByeEsFlWkm8ScEII8yiKYjV3MljHuV4hRKNSXw+8LCoqIjo6mqFDhzJs2DCOHz9OYWEh48ePZ8iQIYwfP56rV6uerKyqKkuWLCEoKIgRI0aQmZlpdP8ScEIIsymKaS9jYmJiGDBgAF9//TVJSUl06NCB2NhY/P392bNnD/7+/sTGxgJVk1hduHCBPXv2sHjxYhYsWGB0/xJwQgiz1UcPrri4mPT0dMaMGQOAk5MT7u7uJCcnEx4eDkB4eDj79u0DMCxXFIUePXpQVFREbm5urW3IMTghhHnMuNA3Pz+fiRMnGt5HRkYSGRkJwMWLF9FoNLz22mucOXMGPz8/5s6dS15enmHGei8vL/Lyqp6+/N/zvPj4+KDT6Wqd3V4CTghhFgXTH5ek0WiIj4+/47qKigpOnTrFvHnz6N69O0uWLDEMRw1t1XHyGhmiCiHMVh9DVB8fH3x8fOjevWoazaFDh3Lq1Ck8PT0NQ8/c3Fw0Gg1w+zwvOTk5Rud5kYATQpipfqYN9PLywsfHh59//hmAQ4cO0aFDBwIDA0lMTAQgMTGRwYOrZqS7tVxVVTIyMnBzc6t1eAoyRBVCmKseb7afN28er7zyCuXl5fj6+vLWW29RWVnJtGnT2Lp1K23atGHVqlVA1RQIKSkpBAUF4eLiwtKlS43uXwJOCGGW+rzZvmvXrnc8Rrd+/frb21UU5s+fb9b+JeCEEGaTW7WEEDbLWm7VkoATQphFUcDOSrpwEnBCCLNZSb5JwAkhzCfPgxNC2CwrOQRXc8AtXry41pR+4403GqQgIUTjZk3Pg6sx4B555JF7WYcQwoooWHnARUREVHt/48YNXFxcGrwgIUTjZyUdOOP3oh4/fpzhw4czbNgwAM6cOWPSg+aEELbp1p0M9fFE34ZmNOCWLl3KRx99RKtWrQDo0qULR48ebfDChBCNV3090behmXQWtXXr1tXe29nJQ0iEaMps5kLf1q1bc+zYMRRFoby8nA0bNtChQ4d7UZsQohFSFOu5VctoV2zBggVs2rQJnU7HgAEDOH36NG+++ea9qE0I0UjZzBBVo9Hwzjvv3ItahBBWQbGaIarRHlxWVhaTJ0+mX79++Pv78+KLL5KVlXUvahNCNFKKiS9LMxpwM2fOZOjQoXz//fd89913DB06lBkzZtyL2oQQjZBNXSZy48YNwsPDcXBwwMHBgbCwMEpLS+9FbUKIxkiputDXlJel1XgMrrCwEICBAwcSGxvL8OHDURSFXbt2ERAQcM8KFEI0PtZyFrXGgBs1ahSKoqCqKgCbN282rFMUhZkzZzZ8dUKIRqc+52QIDAykefPm2NnZYW9vT3x8PIWFhUyfPp1Lly5x//33s2rVKlq2bImqqsTExJCSkkKzZs1YtmwZfn5+te6/xoDbv39/vXwBIYTtqc8O3Pr16w1znwLExsbi7+/PpEmTiI2NJTY2llmzZpGamsqFCxfYs2cPJ06cYMGCBWzZsqXWfZt0J8NPP/3E2bNnKSsrMywLDw+/y68jhLBqSsM+8DI5OZmNGzcCVTkTFRXFrFmzSE5OJjw8HEVR6NGjB0VFReTm5tY6N6rRgFu7di1paWmcO3eOgIAAUlNT6dWrlwScEE1UfV8C8vvf/x5FUYiMjCQyMpK8vDxDaHl5eZGXlweATqfDx8fH8DkfHx90Ol3dAm737t0kJSURHh7OW2+9xZUrV5g1a1Zdv5MQwmop2Js4Rs3Pz2fixImG97dC7JbPP/8crVZLXl4e48ePp3379tVbquPlJkYDztnZGTs7OxwcHLh27Rqenp5kZ2ffdYNCCOtnauhoNJo7Tux8i1arBcDT05OgoCBOnjyJp6enYeiZm5trOD6n1WrJyckxfDYnJ8fw+ZoYvQ7ukUceoaioiLFjxzJq1CgiIiLo2bOnSV9OCGF7qs6i1v1e1OvXr3Pt2jXDnw8cOECnTp0IDAwkMTERgMTERAYPHgxgWK6qKhkZGbi5udU6PAUTenC3Hm759NNPM2DAAK5du0aXLl2MfUwIYavqaV7UvLw8pkyZAoBeryc0NJSBAwfSrVs3pk2bxtatW2nTpg2rVq0CICAggJSUFIKCgnBxcWHp0qVG26gx4DIzM2v8UGZmptHrT4QQtqs+TqL6+vqybdu225Z7eHiwfv36O7SpMH/+fLPaqDHgli1bVuOHFEVhw4YNZjVkip5dH6AgfW2971c0nMQfLlm6BGGGZg7187DaxnCfqSlqDLhb16EIIcRvKYC9tQecEELcUSO5kd4UEnBCCLMoSMAJIWxW43jWmymMHnFUVZWkpCTWrq06+H/58mVOnjzZ4IUJIRova3kenEmTzmRkZLBz504AmjdvzsKFCxu8MCFE41RfF/reC0YD7uTJk8yfPx9nZ2cAWrZsSXl5eYMXJoRopBRwUBSTXpZm9Bicg4MDer3eMObOz8+XiZ+FaMJu9eCsgdGAi4qKYsqUKeTl5bFy5Uq+/vprpk2bdi9qE0I0UtYybaDRgBs5ciR+fn4cPnwYVVV57733ZGZ7IZo4K8k34wF3+fJlXFxcGDRoULVlbdq0adDChBCNk01dB/fCCy8Y/lxaWsrFixd56KGHDGdVhRBNjILJD7y0NKMBt3379mrvMzMz+eyzzxqsICFE42ZTPbj/5ufnJxf6CtHEKfU6K0PDMRpwcXFxhj9XVlZy6tQpo0/RFELYNpvpwZWUlBj+bG9vT0BAAMHBwQ1alBCi8bKZIaper6ekpIRXX331XtUjhGjs6jjT1b1UY8BVVFTg4ODAsWPH7mU9QggrYG8lNzPVGHBjx44lISGBLl26MHnyZIYOHYqrq6th/ZAhQ+5JgUKIxqVqiGrlPbhbysrK8PDwIC0trdpyCTghmq76PAan1+sZPXo0Wq2WDz74gKysLGbMmEFhYSF+fn4sX74cJycnysrKmD17NpmZmbRq1YqVK1fStm3b2uusaUVeXh5xcXF06tSJzp0706lTJ8Orc+fO9ffthBDWxcRHJZnayduwYUO12z9XrFjBuHHj2Lt3L+7u7mzduhWALVu24O7uzt69exk3bhwrVqwwuu8aA66yspKSkhKuX79u+P+t12/PrAohmhYFsEMx6WVMTk4O3377LWPGjAGqHrB7+PBhw5UaERERJCcnA7B//34iIiIACA4O5tChQ6iqWuv+axyienl58cc//tGkLyyEaFpM7Z3l5+czceJEw/vIyEgiIyMN75cuXcqsWbMMnaaCggLc3d1xcKiKJh8fH3Q6HQA6nY7WrVsDVY9xc3Nzo6CgAI1GU2P7NQacsWQUQjRNCuBg4kE4jUZDfHz8Hdd98803aDQaHnnkkduO8deXGgPuk08+aZAGhRDWrz5Ooh47doz9+/eTmppKaWkp165dIyYmhqKiIsNlajk5OWi1WgC0Wi3Z2dn4+PhQUVFBcXExHh4etbZR4zG4Vq1a1f0bCCFsj1J1mYgpr9rMnDmT1NRU9u/fz5///Gf69evHO++8Q9++fdm9ezcACQkJBAYGAhAYGEhCQgIAu3fvpl+/fkYvOLaSy/WEEI1FQ086M2vWLOLi4ggKCqKwsJCxY8cCMGbMGAoLCwkKCiIuLo5XXnnFeK1qIzrYVqlCmd7SVQhzJP5wydIlCDMMedgLjatTnfZxpaSUHad0Jm07rs8DdWqrrmTiZyGEmYwPPxsLCTghhFls6lYtIYT4b9YRbxJwQoi7YCUdOAk4IYSZFKz/eXBCCHEnVfeiWgcJOCGE2eQkgxDCZskQVQhhk2SIKoSwYTYw6YwQQtTEOuJNAk4IYSYFsJcenBDCJtXhSSH3mgScEMJsipUMUiXghBBmufU8OGsgASeEMJspM2Y1BhJwQgizSQ9OCGGT5HlwQgibZuKsgRYnASeEMJMiZ1GFEDaqnq6DKy0t5ZlnnqGsrAy9Xk9wcDDR0dFkZWUxY8YMCgsL8fPzY/ny5Tg5OVFWVsbs2bPJzMykVatWrFy5krZt29bahrXcM9uovTBxAg+08aZXj0cMy/Lz8wkZGsQjXTsRMjSIgoICALZvS6JPz/+lb68e/K5vbw58/72lym6yykpv8kZUCK9GBvHKmEC2vL8CAFVV+WLtn5gePoCZo57g688/qva5c5kZPNOnHWn7dlii7EZD4VYfzvh/tXFycmL9+vVs27aNxMREvvvuOzIyMlixYgXjxo1j7969uLu7s3XrVgC2bNmCu7s7e/fuZdy4caxYscJorQ0WcK+99hr+/v6EhoY2VBONRtTz40ja8XW1ZSuWL+OJwMH84/Q/eSJwMCuWLwNgUOBgjhw7QdrfM/jrhx/z0uSJlii5SXN0cuaND/7Gn77Yy7LPd3Pi0Lf88+TfSdn2N/J0l3knPoV34r/FPzjM8JlKvZ7P/rKU/+030IKVNx52immv2iiKQvPmzQGoqKigoqICRVE4fPgwwcHBAERERJCcnAzA/v37iYiIACA4OJhDhw5hbNbTBgu4UaNGsW7duobafaPSf8BANBpNtWU7tifxbNTzADwb9TzbtyUC0KJFC8OTGEpKSqzmqQy2RFEUmrlW/WLpKyrQ//sXa9/WDYz6wzTs7Kp+LVpq7jN85uvNcfQdPBz33yxrykyd2T4/P59Ro0YZXl988UW1/ej1esLCwnj88cd5/PHH8fX1xd3dHQeHqqNnPj4+6HRVc7DqdDpat24NgIODA25uboaRUU0a7Bhcnz59uHjxYkPtvtHL/c1fho+PD7m6/0yUm5SYwJtvvMa/cnOJT9ppqRKbtEq9ntefGUZO1gWG/N/zdOz2KLqLv3Boz3bSv/kadw8Nz89eROsH2pOfm036N18xL3YL5xbOtHTpFqdg+tNENBoN8fHxNa63t7cnKSmJoqIipkyZws8//1wvNd4ix+DuAUWp/vyssPAITvzjDH/7MpFFC+ZZsLKmy87enmWb9/Du1+mcy8wg6+wZysvKcHR2ZummXQRG/D8+WPAKABtWLOD/Rb9u6NkJ03twpnJ3d6dv375kZGRQVFRERUUFADk5OWi1WgC0Wi3Z2dlA1ZC2uLgYDw+P2uu8y+8njPD+zV9GdnY2Xt7et23Tf8BAzp//mStXrtzr8sS/NXdryf/0fpwTB7/FU9uaxwKHAdAncBi/nj0NwM+nTmB8ZOcAAAsySURBVLL6tSlMDelH2r6dfPzWXNK/+bqWvdo+xcRXbfLz8ykqKgLg5s2bHDx4kA4dOtC3b192794NQEJCAoGBgQAEBgaSkJAAwO7du+nXr5/RQzxymUgDCQkdyacb1zNr9hw+3bie0BFVB6zPnT1L+w4dUBSF48eOUVpaiqenp4WrbVqKCvKwd3CguVtLym7e4IfD3zFy3Ev0fiKYzPSDeN//AKf/fojWD7QHYPWOQ4bPvj9/Oo8OGEyfQUMtVX7jUA+HjnNzc5kzZw56vR5VVRk6dCiDBg2iY8eOTJ8+nVWrVtG1a1fGjh0LwJgxY5g1axZBQUG0bNmSlStXGm1DAq4ePPfs03yX8i1Xrlyhw4NtmffmQl6ZPYdnn/4/1sd9xAMPtOPTz/8GQELCl3z26QYcHRxp5uLCxk1fyImGe6zgXzrenz+dyn//YvULCuXRgU/ycM8+rJ07la8++5BmLs2Z9Obbli61UVKU+rlVq0uXLiQmJt623NfX13BpyG85OzuzevVqs9pQVGPnWe/SjBkzOHLkCAUFBXh6ejJ16lRDEtekUoUyfUNUIxpK4g+XLF2CMMOQh73QuDrVaR8lpRWculRi0rZ92resU1t11WA9uD//+c8NtWshhKVZyaBDhqhCCDPJvahCCBtmLYeNJeCEEGazknyTgBNCmKdqTgbriDgJOCGE2awk3yTghBDms5J8k4ATQpjJnLvtLUwCTghhNrlMRAhhs+QYnBDCJsnM9kIImyZDVCGEzZIenBDCZllJvknACSHugpUknAScEMIs9fXAy3tBAk4IYTbriDcJOCHE3bCShJOAE0KYyXoeeCnTBgohzKYopr1qk52dTVRUFMOHDyckJIT169cDUFhYyPjx4xkyZAjjx4/n6tWrAKiqypIlSwgKCmLEiBFkZmYarVMCTghhtvqYF9Xe3p45c+awa9cuvvjiCz777DPOnj1LbGws/v7+7NmzB39/f2JjYwFITU3lwoUL7Nmzh8WLF7NgwQKjdUrACSHMcuuBl6a8auPt7Y2fnx8ALVq0oH379uh0OpKTkwkPDwcgPDycffv2ARiWK4pCjx49KCoqIjc3t9Y25BicEMJspl4lkp+fz8SJEw3vIyMjiYyMvG27ixcvcvr0abp3705eXh7e3t4AeHl5kZeXB4BOp8PHx8fwGR8fH3Q6nWHbO5GAE0KYzdRTDBqNhvj4+Fq3KSkpITo6mtdff50WLVpUb8eEnmBtZIgqhDCPqQfgTMil8vJyoqOjGTFiBEOGDAHA09PTMPTMzc1Fo9EAoNVqycnJMXw2JycHrVZb6/4l4IQQZlNM/K82qqoyd+5c2rdvz/jx4w3LAwMDSUxMBCAxMZHBgwdXW66qKhkZGbi5udU6PAUZogoh7kJ93Kn197//naSkJDp37kxYWBgAM2bMYNKkSUybNo2tW7fSpk0bVq1aBUBAQAApKSkEBQXh4uLC0qVLjdepqqpa91LrR6UKZXpLVyHMkfjDJUuXIMww5GEvNK5OddpHWUUlucXlJm3b1sO5Tm3VlfTghBB3wTruZJCAE0KYzUoeJiIBJ4Qwn5XkmwScEMJ80oMTQtikqhvprSPhJOCEEGazjniTgBNC3AUr6cBJwAkhzGU9D7yUgBNCmM868k0CTghhPivJNwk4IYR5FGTaQCGEDbOSfJPHJQkhbJf04IQQ5jFhxqzGQgJOCGE2uUxECGGTqmbVsnQVppGAE0KYTQJOCGGzZIgqhLBZ0oMTQtgsK8k3CTghxF2wkoSTgBNCmEVRFOysJOAa1bSBQghRn+RWLSGEzZKAE0LYLAk4IYTNkoATQtgsCTghhM2SgBNC2CwJOCGEzZKAa0CpqakEBwcTFBREbGyspcsRJnjttdfw9/cnNDTU0qWIeiAB10D0ej2LFi1i3bp17Ny5kx07dnD27FlLlyWMGDVqFOvWrbN0GaKeSMA1kJMnT9KuXTt8fX1xcnIiJCSE5ORkS5cljOjTpw8tW7a0dBminkjANRCdToePj4/hvVarRafTWbAiIZoeCTghhM2SgGsgWq2WnJwcw3udTodWq7VgRUI0PRJwDaRbt25cuHCBrKwsysrK2LlzJ4GBgZYuS4gmRR6X1IBSUlJYunQper2e0aNH8+KLL1q6JGHEjBkzOHLkCAUFBXh6ejJ16lTGjh1r6bLEXZKAE0LYLBmiCiFslgScEMJmScAJIWyWBJwQwmZJwAkhbJYEnBXp2rUrYWFhhIaGEh0dzY0bN+56X3PmzOHrr78GYO7cubU+CCAtLY1jx46Z3UZgYCD5+fkmL/+tnj17mtXWmjVr+Oijj8z6jLB9EnBWpFmzZiQlJbFjxw4cHR3ZvHlztfUVFRV3td+YmBg6duxY4/ojR45w/Pjxu9q3EJYkEz9bqd69e/Pjjz+SlpbGX/7yF9zd3Tl//jy7du1ixYoVHDlyhLKyMp555hmeeuopVFVl8eLFHDhwgNatW+Po6GjYV1RUFLNnz6Zbt26kpqaycuVK9Ho9Hh4exMTEsHnzZuzs7Ni2bRvz5s2jffv2zJ8/n8uXLwPw+uuv06tXLwoKCpg5cyY6nY4ePXpgyiWWL730Ejk5OZSWlvLcc88RGRlpWLd06VIOHDjAfffdx8qVK9FoNPz6668sXLiQgoICmjVrxuLFi+nQoUP9/4CFbVCF1ejRo4eqqqpaXl6uTp48Wd20aZN6+PBhtXv37uqvv/6qqqqqbt68WX333XdVVVXV0tJSNSIiQv3111/V3bt3q+PGjVMrKirUnJwctVevXupXX32lqqqqPvvss+rJkyfVvLw8deDAgYZ9FRQUqKqqqqtXr1bXrVtnqGPGjBlqenq6qqqqeunSJXXo0KGqqqrq4sWL1TVr1qiqqqrffPON2rlzZzUvL++27zFo0CDD8ltt3LhxQw0JCVHz8/NVVVXVzp07q0lJSaqqquqaNWvUhQsXqqqqqs8995x6/vx5VVVVNSMjQ42KirpjjUKoqqpKD86K3Lx5k7CwMKCqBzdmzBiOHz9Ot27d8PX1BeDAgQP8+OOP7N69G4Di4mJ++eUX0tPTCQkJwd7eHq1WS79+/W7bf0ZGBr179zbsq1WrVnes4+DBg9WO2V27do2SkhLS09NZu3YtAE888YRJz1XbuHEje/fuBSA7O5tffvkFDw8P7OzsGD58OABhYWH88Y9/pKSkhOPHj/Pyyy8bPl9WVma0DdF0ScBZkVvH4P6bq6ur4c+qqvLGG28wYMCAatukpKTUWx2VlZX87W9/w9nZuU77SUtL4+DBg3zxxRe4uLgQFRVFaWnpHbdVFAVVVXF3d7/jz0CIO5GTDDamf//+fP7555SXlwNw/vx5rl+/Tp8+ffjqq6/Q6/Xk5uaSlpZ222d79OjB0aNHycrKAqCwsBCA5s2bU1JSUq2NjRs3Gt6fPn0aqHoa7vbt24GqQL169WqttRYXF9OyZUtcXFw4d+4cGRkZhnWVlZWGXuj27dvp1asXLVq0oG3btnz11VdAVZifOXPGvB+QaFIk4GzM2LFj6dixI6NGjSI0NJQ333wTvV5PUFAQ7dq1Y/jw4bz66qv06NHjts9qNBoWLVrE1KlTGTlyJNOnTwdg0KBB7N27l7CwMI4ePcrcuXP5xz/+wYgRIxg+fDiff/45AFOmTOHo0aOEhISwd+9e2rRpU2utAwcOpKKigmHDhvHOO+9Uq8nV1ZWTJ08SGhrK4cOHmTJlCgBvv/02W7duZeTIkYSEhLBv3776+tEJGyRPExFC2CzpwQkhbJYEnBDCZknACSFslgScEMJmScAJIWyWBJwQwmZJwAkhbNb/B88Fp80Cw0iOAAAAAElFTkSuQmCC\n",
            "text/plain": [
              "<Figure size 432x288 with 2 Axes>"
            ]
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "tPK249v8S8Xs"
      },
      "source": [
        "### Otimizando Hiperparâmetros XGBoost (XGB)\n",
        "\n",
        "XGBoost possui diversos parâmetros, alguns que afetam ainda mais o desempenho do modelo e outros que nem tanto. O recomendado é começar definindo uma taxa de aprendizado e o número de estimadores, começar a procura por outros parâmetros e depois checar outras taxas de aprendizado.\n",
        "\n",
        "Portanto, vamos começar a procura pelo `n_estimators` com o `learning_rate` instanciado em 0.01"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "nnWHLOYkB0QV",
        "outputId": "f0abe9d8-4785-4754-b1ad-bc2ee530a178"
      },
      "source": [
        "# Modelo a ser otimizado\n",
        "xgb_rus_Rscaled = XGBClassifier(learning_rate=0.01)\n",
        "\n",
        "# Parâmetros a serem otimizados\n",
        "param_grid = {\n",
        "    'n_estimators':[100, 200, 1000],\n",
        "}\n",
        "\n",
        "# Identificar melhores parâmetros\n",
        "kfold = StratifiedKFold(n_splits=10, shuffle=True)\n",
        "grid_search = GridSearchCV(xgb_rus_Rscaled, param_grid, scoring='recall', n_jobs=-1, cv=kfold)\n",
        "grid_result = grid_search.fit(X_rus_Rscaled, y_train_rus_Rscaled)\n",
        "\n",
        "# Ver Resultados\n",
        "print(f'Melhor: {grid_result.best_score_} para {grid_result.best_params_}')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Melhor: 0.8380851063829787 para {'n_estimators': 200}\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "oIYLLOOUrD34"
      },
      "source": [
        "Uma vez com o número de estimadores definido para 200, vou realizar a busca para os parâmetros `max_depth` e `min_child_weight.`"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "g1t9MCxHFrNW",
        "outputId": "125efea6-c61a-4e68-9081-b22977505f7d"
      },
      "source": [
        "# Modelo a ser otimizado\n",
        "xgb_rus_Rscaled = XGBClassifier(learning_rate=0.01, n_estimators=200)\n",
        "\n",
        "# Parâmetros a serem otimizados\n",
        "param_grid = {\n",
        "    'max_depth':[1, 3, 6],\n",
        "    'min_child_weight':[1, 5, 10]\n",
        "}\n",
        "\n",
        "# Identificar melhores parâmetros\n",
        "kfold = StratifiedKFold(n_splits=10, shuffle=True)\n",
        "grid_search = GridSearchCV(xgb_rus_Rscaled, param_grid, scoring='recall', n_jobs=-1, cv=kfold)\n",
        "grid_result = grid_search.fit(X_rus_Rscaled, y_train_rus_Rscaled)\n",
        "\n",
        "# Ver resultados\n",
        "print(f'Melhor: {grid_result.best_score_} para {grid_result.best_params_}')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Melhor: 0.8573606889564337 para {'max_depth': 1, 'min_child_weitght': 1}\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "jes9v6s9rpcn"
      },
      "source": [
        "Após obter os valores `max_depth=1` e `min_child_weight=1`, é necessário começar a otimização do parâmetro `gamma`"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "e4G1NSU9G37P",
        "outputId": "8588ac15-e8d8-47da-ff06-df8ea5308890"
      },
      "source": [
        "# Modelo a ser otimizado\n",
        "xgb_rus_Rscaled = XGBClassifier(learning_rate=0.01, n_estimators=200, max_depth=1, min_child_weight=1)\n",
        "\n",
        "# Parâmetros a serem otimizados\n",
        "param_grid = {\n",
        "    'gamma':[0, 1, 5]\n",
        "}\n",
        "\n",
        "# Identificar melhores parâmetros \n",
        "kfold = StratifiedKFold(n_splits=10, shuffle=True)\n",
        "grid_search = GridSearchCV(xgb_rus_Rscaled, param_grid, scoring='recall', n_jobs=-1, cv=kfold)\n",
        "grid_result = grid_search.fit(X_rus_Rscaled, y_train_rus_Rscaled)\n",
        "\n",
        "# Ver resultados\n",
        "print(f'Melhor: {grid_result.best_score_} para {grid_result.best_params_}')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Melhor: 0.8573353596757853 para {'gamma': 0}\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "5F8W3i-isANj"
      },
      "source": [
        "Com `gamma=0`, chegou a hora de testar outros valores para `learning_rate`. Dessa forma, podemos otimizar ainda mais o modelo e saber qual valor nos entregará o melhor resultado."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "n64Z1pRwHq_p",
        "outputId": "32b94cfb-d6a5-43bf-dde5-da665a876643"
      },
      "source": [
        "# Modelo a ser otimizado\n",
        "xgb_rus_Rscaled = XGBClassifier(n_estimators=200, max_depth=1, min_child_weight=1, gamma=0)\n",
        "\n",
        "# Parâmetros a serem otimizados\n",
        "param_grid = {\n",
        "    'learnig_rate':[0.01, 0.015, 0.1]\n",
        "}\n",
        "\n",
        "# Identificar melhores parâmetros \n",
        "kfold = StratifiedKFold(n_splits=10, shuffle=True)\n",
        "grid_search = GridSearchCV(xgb_rus_Rscaled, param_grid, scoring='recall', n_jobs=-1, cv=kfold)\n",
        "grid_result = grid_search.fit(X_rus_Rscaled, y_train_rus_Rscaled)\n",
        "\n",
        "# Ver resultados\n",
        "print(f'Melhor: {grid_result.best_score_} para {grid_result.best_params_}')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Melhor: 0.8095542046605877 para {'learnig_rate': 0.01}\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "yqHdHcdeTFcg"
      },
      "source": [
        "#### Avaliando o Desempenho do Modelo XGB\n",
        "\n",
        "De acordo com a cross-validation, definir um valor específico para a taxa de aprendizagem, fez com que o desempenho do modelo caísse, quando levamos em conta a métrica `recall` mas acredito que o cenário não será esse quando o modelo fizer as previsões.\n",
        "\n",
        "Agora vou treinar o modelo em cima dos dados de treino, após o treino podemos verificar o desemepenho do modelo com base nas previsões feitas nos dados de teste, dados que o modelo não teve nenhum contato até agora."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 482
        },
        "id": "Rt4qqAoXIZgx",
        "outputId": "0e12fed9-da26-47b4-defd-66c25c4fc7e8"
      },
      "source": [
        "# Modelo final\n",
        "xgb_rus_Rscaled = XGBClassifier(learning_rate=0.01 ,n_estimators=200, max_depth=1, min_child_weight=1, gamma=0)\n",
        "\n",
        "# Treinar Modelo\n",
        "xgb_rus_Rscaled.fit(X_rus_Rscaled, y_train_rus_Rscaled)\n",
        "\n",
        "# Fazer previsões em cima dos dados de teste\n",
        "y_pred_XgbRus = xgb_rus_Rscaled.predict(X_test_Rscaled)\n",
        "\n",
        "# Plotar matriz de confusão\n",
        "skplt.metrics.plot_confusion_matrix(y_test, y_pred_XgbRus)\n",
        "\n",
        "# Imprimir Relatório de Classificação\n",
        "print(classification_report(y_test, y_pred_XgbRus, digits=4))\n",
        "\n",
        "# Imprimir área sob curva\n",
        "print('AUC: {:.4f}\\n'.format(roc_auc_score(y_test, y_pred_XgbRus)))\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "              precision    recall  f1-score   support\n",
            "\n",
            "           0     0.9105    0.6368    0.7494      1294\n",
            "           1     0.4509    0.8266    0.5835       467\n",
            "\n",
            "    accuracy                         0.6871      1761\n",
            "   macro avg     0.6807    0.7317    0.6665      1761\n",
            "weighted avg     0.7886    0.6871    0.7054      1761\n",
            "\n",
            "AUC: 0.7317\n",
            "\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 2 Axes>"
            ]
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "tuItBaSnUgzo"
      },
      "source": [
        "## Método Ensemble para os Modelos do Balanceamento RUS\n",
        "\n",
        "A fim de obter resultados mais consistentes e um desempenho ainda melhor, vou usar um método ensemble, que tem como objetivo combinar as previsões de diversos modelos. Para isso, vou usar os 3 modelos que foram trabalhados acima.\n",
        "\n",
        "O método ensemble utilizado será o `VotingClassifier`, a previsão final que ele nos dá é a que ocorre com maior frequência.\n",
        "\n",
        "Abaixo é possível ver que o `VotingClassifier` possui a maior pontuação de Recall dentre os 4 modelos, porém é apenas pouca coisa maior que a a pontuação do modelo `LDA`."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 173
        },
        "id": "zfyaKz6bgaUk",
        "outputId": "20cdac08-7af9-43ca-fdd5-c45a63fd4e7d"
      },
      "source": [
        "# Modelos que vão compor o Voting CLassifier\n",
        "xgb_ensemble_rus_Rscaled = XGBClassifier(learning_rate=0.01,n_estimators=200, max_depth=1, min_child_weight=1, gamma=0)\n",
        "lda_ensemble_rus_Rscaled = LinearDiscriminantAnalysis(solver='lsqr', shrinkage='auto') \n",
        "svc_ensemble_rus_Rscaled = SVC(C=0.01, kernel='poly', degree=3, gamma=1, probability=True)\n",
        "voting_clf_rus = VotingClassifier(estimators=[('XGB', xgb_ensemble_rus_Rscaled), ('LDA', lda_ensemble_rus_Rscaled), ('SVC', svc_ensemble_rus_Rscaled)])\n",
        "\n",
        "# realizar treino dos modelos\n",
        "for model in (xgb_ensemble_rus_Rscaled, lda_ensemble_rus_Rscaled, svc_ensemble_rus_Rscaled, voting_clf_rus):\n",
        "  model.fit(X_rus_Rscaled, y_train_rus_Rscaled)\n",
        "\n",
        "model = []\n",
        "Recall = []\n",
        "\n",
        "# Mostrar o recall com base nas previsões dos modelos e do Voting Classifier\n",
        "for clf in (xgb_ensemble_rus_Rscaled, lda_ensemble_rus_Rscaled, svc_ensemble_rus_Rscaled, voting_clf_rus):\n",
        "  y_pred_ensemble = clf.predict(X_test_Rscaled)\n",
        "  model.append(clf.__class__.__name__)\n",
        "  Recall.append(recall_score(y_test, y_pred_ensemble))\n",
        "\n",
        "\n",
        "pd.DataFrame(data=Recall, index=model, columns=[\"Recall\"])\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>Recall</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>XGBClassifier</th>\n",
              "      <td>0.826552</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>LinearDiscriminantAnalysis</th>\n",
              "      <td>0.766595</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>SVC</th>\n",
              "      <td>0.779443</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>VotingClassifier</th>\n",
              "      <td>0.785867</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "                              Recall\n",
              "XGBClassifier               0.826552\n",
              "LinearDiscriminantAnalysis  0.766595\n",
              "SVC                         0.779443\n",
              "VotingClassifier            0.785867"
            ]
          },
          "metadata": {},
          "execution_count": 15
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "toOfn4X1UsZC"
      },
      "source": [
        "### Otimizando Hiperparâmetros Voting Classifier\n",
        "\n",
        "O `VotingClassifier` possui uma quantidade pequena de parâmetros, sendo necessário especificar apenas o parâmetros `estimators`, onde serão passados os estimadores que serão utilizados. Porém, irei buscar a otimização de outros dois parâmetros, são eles:\n",
        "* `Voting` - tendo como possibilidade os seguintes argumentos `hard` e `soft`\n",
        "* `n_jobs` - esse é o número de trabalhos realizados para o classificador se adequar."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "WTRM9Zd6MyTB",
        "outputId": "7e840e91-992c-4b54-d277-24e76e27e6a2"
      },
      "source": [
        "# Modelo a ser otimizado\n",
        "voting_clf_rus = VotingClassifier(estimators=[('XGB', xgb_ensemble_rus_Rscaled), ('LDA', lda_ensemble_rus_Rscaled), ('SVC', svc_ensemble_rus_Rscaled)])\n",
        "\n",
        "# Parâmetros a serem otimizados\n",
        "param_grid = {\n",
        "      'voting':('hard', 'soft')\n",
        "}\n",
        "\n",
        "# Identificar melhores parâmetros \n",
        "kfold = StratifiedKFold(n_splits=10, shuffle=True)\n",
        "grid_search = GridSearchCV(voting_clf_rus, param_grid, scoring='recall', n_jobs=-1, cv=kfold)\n",
        "grid_result = grid_search.fit(X_rus_Rscaled, y_train_rus_Rscaled)\n",
        "\n",
        "# Ver resultados\n",
        "print(f'Melhor: {grid_result.best_score_} para {grid_result.best_params_}')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Melhor: 0.8245288753799391 para {'voting': 'hard'}\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "E0tyZBot5s6F"
      },
      "source": [
        "Após a otimização do parâmetro `voting = hard` foi possível perceber uma melhora na métrica `recall` que subiu de 0.74 para 0.80. O próximo passo é  otimizar o parâmetro `n_jobs`, a fim de saber qual valor nos entregará o melhor desemepenho do classificador."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "ypmvmbQ4Nm03",
        "outputId": "8d09d299-d278-4c90-9ed1-6683721e5128"
      },
      "source": [
        "# Modelo a ser otimizado\n",
        "voting_clf_rus = VotingClassifier(estimators=[('XGB', xgb_ensemble_rus_Rscaled), ('LDA', lda_ensemble_rus_Rscaled), ('SVC', svc_ensemble_rus_Rscaled)]\n",
        "                              , voting='hard')\n",
        "# Parâmetros a serem otimizados\n",
        "param_grid = {\n",
        "        'n_jobs':range(-1, 3, 4)\n",
        "}\n",
        "\n",
        "# Identificar melhores parâmetros \n",
        "kfold = StratifiedKFold(n_splits=10, shuffle=True)\n",
        "grid_search = GridSearchCV(voting_clf_rus, param_grid, scoring='recall', n_jobs=-1, cv=kfold)\n",
        "grid_result = grid_search.fit(X_rus_Rscaled, y_train_rus_Rscaled)\n",
        "\n",
        "# Ver resultados\n",
        "print(f'Melhor: {grid_result.best_score_} para {grid_result.best_params_}')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Melhor: 0.8116919959473151 para {'n_jobs': -1}\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "5OYWsjjVU_Dm"
      },
      "source": [
        "#### Avaliando o Desempenho do Voting Classifier\n",
        "\n",
        "Tendo otimizados os parâmetros do nosso `VotingClassifier`, vamos realizar sua avaliação de desempenho, agora entrando em contato com os dados de teste. \n",
        "\n",
        "É importante ficar de olho na coluna **recall**, referente ao **churn = 1**, ela nos mostra a quantidade de previsões certas quando o churn ocorre. \n",
        "\n",
        "Portanto, se fizermos uma comparação do desempenho individual de cada um dos modelos usados para compor o método ensemble, nota-se que o melhor desempenho foi o do **XGBoost** que obteve uma métrica do **recall = 0.7317** e preveu corretamente **349**, já o **VotingClassifier** obteve um valor um pouco melhor com o **recall = 0.748** e preveu corretamente, apenas um churn a mais, sendo **350**."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 482
        },
        "id": "TdIMGokPOV5-",
        "outputId": "a6499f7b-0522-4da7-c7ae-bbb706bd7560"
      },
      "source": [
        "# Modelo final\n",
        "voting_clf_rus = VotingClassifier(estimators=[('XGB', xgb_ensemble_rus_Rscaled), ('LDA', lda_ensemble_rus_Rscaled), ('SVC', svc_ensemble_rus_Rscaled)]\n",
        "                              , voting='hard', n_jobs=-1)\n",
        "\n",
        "# Treinar Modelo\n",
        "voting_clf_rus.fit(X_rus_Rscaled, y_train_rus_Rscaled)\n",
        "\n",
        "# Fazer previsões em cima dos dados de teste\n",
        "y_pred_voting_clf_rus = voting_clf_rus.predict(X_test_Rscaled)\n",
        "\n",
        "# Plotar matriz de confusão\n",
        "skplt.metrics.plot_confusion_matrix(y_test, y_pred_voting_clf_rus)\n",
        "\n",
        "# Imprimir Relatório de Classificação\n",
        "print(classification_report(y_test, y_pred_voting_clf_rus, digits=4))\n",
        "\n",
        "# Imprimir área sob curva\n",
        "print('AUC: {:.4f}\\n'.format(roc_auc_score(y_test, y_pred_voting_clf_rus)))\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "              precision    recall  f1-score   support\n",
            "\n",
            "           0     0.8920    0.7465    0.8128      1294\n",
            "           1     0.5162    0.7495    0.6114       467\n",
            "\n",
            "    accuracy                         0.7473      1761\n",
            "   macro avg     0.7041    0.7480    0.7121      1761\n",
            "weighted avg     0.7923    0.7473    0.7594      1761\n",
            "\n",
            "AUC: 0.7480\n",
            "\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 2 Axes>"
            ]
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "xstxntEXL8Fu"
      },
      "source": [
        "## Definindo Modelos para o Balanceamento SMOTE\n",
        "\n",
        "Após realizar todo o trabalho com os modelos que usaram o balanceamento RUS vou definir quais serão os 3 modelos usados para o balancemanto **SMOTE** \n",
        "\n",
        "Serão usados os modelos:\n",
        "\n",
        "* **Random Forest (RF)** - com padronização RobustScaler\n",
        "\n",
        "* **XGBoost (XBG)** - com padronização RobustScaler   \n",
        "\n",
        "* **SGD** - com padronização RobustScaler\n",
        "\n",
        "Todos os três modelos passarão por um processo de otimização de seus respectivos hiperparâmetros, a fim de obtermos resultados ainda mais consistentes."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 514
        },
        "id": "YKdU6Dv48ODJ",
        "outputId": "15ab74ce-1f77-49d3-ed2e-9b076f66ad93"
      },
      "source": [
        "# Imprimir DataSet com os Dados de Recall dos Modelos Escolhidos\n",
        "model = []\n",
        "recall= []\n",
        "\n",
        "model.append('Random Forest SmoScaled')\n",
        "recall.append(val_model_balanced(X_smo_scaled, y_train_smo_scaled, rf, quite=True))  \n",
        "model.append('Random Forest SmoRscaled')\n",
        "recall.append(val_model_balanced(X_smo_Rscaled, y_train_smo_Rscaled, rf, quite=True))\n",
        "model.append('Random Forest SmoQTR')\n",
        "recall.append(val_model_balanced(X_smo_qtr, y_train_smo_qtr, rf, quite=True))\n",
        "model.append('Random Forest SmoQTR-N')\n",
        "recall.append(val_model_balanced(X_smo_qtr_n, y_train_smo_qtr_n, rf, quite=True))\n",
        "model.append('Random Forest SmoPWR')\n",
        "recall.append(val_model_balanced(X_smo_pwr, y_train_smo_pwr, rf, quite=True))\n",
        "\n",
        "model.append('XGBoost smoScaled')\n",
        "recall.append(val_model_balanced(X_smo_scaled, y_train_smo_scaled, xgb, quite=True))\n",
        "model.append('XGBoost smoRscaled')\n",
        "recall.append(val_model_balanced(X_smo_Rscaled, y_train_smo_Rscaled, xgb, quite=True))\n",
        "model.append('XGBoost smoQTR')\n",
        "recall.append(val_model_balanced(X_smo_qtr, y_train_smo_qtr, xgb, quite=True))\n",
        "model.append('XGBoost smoQTR-N')\n",
        "recall.append(val_model_balanced(X_smo_qtr_n, y_train_smo_qtr_n, xgb, quite=True))\n",
        "model.append('XGBoost smoPWR')\n",
        "recall.append(val_model_balanced(X_smo_pwr, y_train_smo_pwr, xgb, quite=True))\n",
        "\n",
        "model.append('SGD smoScaled')\n",
        "recall.append(val_model_balanced(X_smo_scaled, y_train_smo_scaled, sgd, quite=True))  \n",
        "model.append('SGD smoRscaled')\n",
        "recall.append(val_model_balanced(X_smo_Rscaled, y_train_smo_Rscaled, sgd, quite=True))\n",
        "model.append('SGD smoQTR')\n",
        "recall.append(val_model_balanced(X_smo_qtr, y_train_smo_qtr, sgd, quite=True))\n",
        "model.append('SGD smoQTR-N')\n",
        "recall.append(val_model_balanced(X_smo_qtr_n, y_train_smo_qtr_n, sgd, quite=True))\n",
        "model.append('SGD smoPWR')\n",
        "recall.append(val_model_balanced(X_smo_pwr, y_train_smo_pwr, sgd, quite=True))\n",
        "\n",
        "pd.DataFrame(data=recall, index=model, columns=['Recall'])"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>Recall</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>Random Forest SmoScaled</th>\n",
              "      <td>0.861856</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Random Forest SmoRscaled</th>\n",
              "      <td>0.867784</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Random Forest SmoQTR</th>\n",
              "      <td>0.868557</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Random Forest SmoQTR-N</th>\n",
              "      <td>0.865722</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Random Forest SmoPWR</th>\n",
              "      <td>0.868041</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>XGBoost smoScaled</th>\n",
              "      <td>0.850515</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>XGBoost smoRscaled</th>\n",
              "      <td>0.853608</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>XGBoost smoQTR</th>\n",
              "      <td>0.856186</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>XGBoost smoQTR-N</th>\n",
              "      <td>0.854639</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>XGBoost smoPWR</th>\n",
              "      <td>0.847423</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>SGD smoScaled</th>\n",
              "      <td>0.796907</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>SGD smoRscaled</th>\n",
              "      <td>0.808505</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>SGD smoQTR</th>\n",
              "      <td>0.783505</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>SGD smoQTR-N</th>\n",
              "      <td>0.796649</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>SGD smoPWR</th>\n",
              "      <td>0.814948</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "                            Recall\n",
              "Random Forest SmoScaled   0.861856\n",
              "Random Forest SmoRscaled  0.867784\n",
              "Random Forest SmoQTR      0.868557\n",
              "Random Forest SmoQTR-N    0.865722\n",
              "Random Forest SmoPWR      0.868041\n",
              "XGBoost smoScaled         0.850515\n",
              "XGBoost smoRscaled        0.853608\n",
              "XGBoost smoQTR            0.856186\n",
              "XGBoost smoQTR-N          0.854639\n",
              "XGBoost smoPWR            0.847423\n",
              "SGD smoScaled             0.796907\n",
              "SGD smoRscaled            0.808505\n",
              "SGD smoQTR                0.783505\n",
              "SGD smoQTR-N              0.796649\n",
              "SGD smoPWR                0.814948"
            ]
          },
          "metadata": {},
          "execution_count": 31
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "v3tFdVE9McD4"
      },
      "source": [
        "### Otimizando Hiperparâmetros do Modelo Random Forest (RF)\n",
        "\n",
        "O algortimo de RandomForest é uma combinação de várias árvores de decisão que usa a média para melhorar a precisão de suas previsões.\n",
        "\n",
        "Esse é um algoritmo que possui diversos parâmetros a serem otimizados, porém irei trabalhar apenas com os que me julgo capaz de otimizar e que aparentam ter mais impacto no desempenho do modelo.\n",
        "\n",
        "Os primeiros parâmetros que irei otimizar, são os seguintes:\n",
        "* `bootstrap` - caso seja falso, todo o conjunto de dados será usado na construção de cada árvore de decisão.\n",
        "\n",
        "* `max_depth` - A profundidade máxima da árvore. Caso não seja especificado os nós são expandidos até que todas as folhas sejam puras ou até que as folhas possuam menos amostras que o `min_samples_split`"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "0mGlidWiJWMD",
        "outputId": "10b51c85-13cf-43b8-96b0-7bc13735aa74"
      },
      "source": [
        "# Modelo a ser otimizado\n",
        "rf_smo_Rscaled = RandomForestClassifier()\n",
        "\n",
        "# Parâmetros a serem otimizados\n",
        "param_grid = {\n",
        "  'bootstrap': [True, False],\n",
        "  'max_depth': [5, 10, 20, 30, 40, 50, None],\n",
        "\n",
        "}\n",
        "\n",
        "# Identificar melhores parâmetros \n",
        "kfold = StratifiedKFold(n_splits=10, shuffle=True)\n",
        "grid_search = GridSearchCV(rf_smo_Rscaled, param_grid, scoring='recall', n_jobs=-1, cv=kfold)\n",
        "grid_result = grid_search.fit(X_smo_Rscaled, y_train_smo_Rscaled)\n",
        "\n",
        "# Ver resultados\n",
        "print(f'Melhor: {grid_result.best_score_} para {grid_result.best_params_}')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Melhor: 0.8984536082474225 para {'bootstrap': False, 'max_depth': 10}\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "NsrW1clcCJgm"
      },
      "source": [
        "Com o `bootstrap=False` e `max_depth=10`, irei realizar a busca pelos parâmetros `max_features` e `min_samples_leaf`.\n",
        "\n",
        "* `max_features` - O número de características a considerar quando se procura a melhor divisão\n",
        "\n",
        "* `min_samples_leaf` - número mínimo de amostras necessárias para estar em um nó de folha\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "M297rPacMhlf",
        "outputId": "19d2efed-7933-4290-f091-342bb5e26961"
      },
      "source": [
        "# Modelo a ser otimizado\n",
        "rf_smo_Rscaled = RandomForestClassifier(bootstrap=False, max_depth=10)\n",
        "\n",
        "# Parâmetros a serem otimizados\n",
        "param_grid = {\n",
        "    'max_features':[2, 3, 5],\n",
        "    'min_samples_leaf':[3, 4, 5]\n",
        "}\n",
        "\n",
        "# Identificar melhores parâmetros \n",
        "kfold = StratifiedKFold(n_splits=10, shuffle=True)\n",
        "grid_search = GridSearchCV(rf_smo_Rscaled, param_grid, scoring='recall', n_jobs=-1, cv=kfold)\n",
        "grid_result = grid_search.fit(X_smo_Rscaled, y_train_smo_Rscaled)\n",
        "\n",
        "# Ver resultados\n",
        "print(f'Melhor: {grid_result.best_score_} para {grid_result.best_params_}')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Melhor: 0.88659793814433 para {'max_features': 3, 'min_samples_leaf': 3}\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "DZ2SbfAt0P3L"
      },
      "source": [
        "É possível notar que o recall vem diminuindo desde a otimização dos primeiros parâmetros, mas por não ser uma diminuição gritante, decidi não considerar e continuar utilizando os valores decididos pelo `GridSearchCV` e `StratifiedKfold`.\n",
        "\n",
        "Portanto, agora irei realizar a busca pelos seguintes parâmetros:\n",
        "\n",
        "* `min_samples_split` - O número mínimo de amostras necessárias para dividir um nó interno\n",
        "\n",
        "* `n_estimators` - O número de árvores na floresta"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "JlRgsofOM2ba",
        "outputId": "21bfc8b2-49dd-4ea5-c45e-78993021b2dd"
      },
      "source": [
        "# Modelo a ser otimizado\n",
        "rf_smo_Rscaled = RandomForestClassifier(bootstrap=False, max_depth=10, max_features=3, min_samples_leaf=3)\n",
        "\n",
        "# Parâmetros a serem otimizados\n",
        "param_grid = {\n",
        "    'min_samples_split':[6, 8, 10, 12],\n",
        "    'n_estimators':[100, 200, 300]\n",
        "}\n",
        "\n",
        "# Identificar melhores parâmetros \n",
        "kfold = StratifiedKFold(n_splits=10, shuffle=True)\n",
        "grid_search = GridSearchCV(rf_smo_Rscaled, param_grid, scoring='recall', n_jobs=-1, cv=kfold)\n",
        "grid_result = grid_search.fit(X_smo_Rscaled, y_train_smo_Rscaled)\n",
        "\n",
        "# Ver resultados\n",
        "print(f'Melhor: {grid_result.best_score_} para {grid_result.best_params_}')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Melhor: 0.8855670103092784 para {'min_samples_split': 6, 'n_estimators': 300}\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "aYEyWHLBMnT4"
      },
      "source": [
        "#### Avaliando o Desempenho do Modelo RF\n",
        "\n",
        "Agora que todos os parâmetros foram otimizados, podemos realizar a avalição de desemepenho do modelo quando em contato com os dados de teste, estes que não entraram em contato com o modelo até o seguinte momento.\n",
        "\n",
        "Como já foi dito anteriormente, a coluna **recall** é de extrema importância para sabermos se o modelo está obtendo um bom desempenho na previsão do churn.\n",
        "\n",
        "Tendo em vista os modelos apresentados até aqui, o modelo de **Random Forest com balanceamento SMOTE** foi o que apresentou o melhor resultado na métrica **recall = 0.7542**.\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 482
        },
        "id": "9NVtKbUyPqUp",
        "outputId": "909433d4-7e79-4a48-c0a1-803bfb2758c6"
      },
      "source": [
        "# Modelo final\n",
        "rf_smo_Rscaled = RandomForestClassifier(bootstrap=False, max_depth=10, max_features=3, min_samples_leaf=3, min_samples_split=6, n_estimators=300)\n",
        "\n",
        "# Treinar Modelo\n",
        "rf_smo_Rscaled.fit(X_smo_Rscaled, y_train_smo_Rscaled)\n",
        "\n",
        "# Fazer previsões em cima dos dados de teste\n",
        "y_pred_rfSmo = rf_smo_Rscaled.predict(X_test_Rscaled)\n",
        "\n",
        "# Plotar matriz de confusão\n",
        "skplt.metrics.plot_confusion_matrix(y_test, y_pred_rfSmo)\n",
        "\n",
        "# Imprimir Relatório de Classificação\n",
        "print(classification_report(y_test, y_pred_rfSmo, digits=4))\n",
        "\n",
        "# Imprimir área sob curva\n",
        "print('AUC: {:.4f}\\n'.format(roc_auc_score(y_test, y_pred_rfSmo)))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "              precision    recall  f1-score   support\n",
            "\n",
            "           0     0.8888    0.7782    0.8298      1294\n",
            "           1     0.5430    0.7302    0.6228       467\n",
            "\n",
            "    accuracy                         0.7655      1761\n",
            "   macro avg     0.7159    0.7542    0.7263      1761\n",
            "weighted avg     0.7971    0.7655    0.7749      1761\n",
            "\n",
            "AUC: 0.7542\n",
            "\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 2 Axes>"
            ]
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Pk0nhkxEM3se"
      },
      "source": [
        "### Otimizando Hiperparâmetros Modelo XGBoost (XGB) \n",
        "\n",
        "O processo aqui é basicamente o mesmo que fiz anteriormente, a principal diferença serão os dados usados.\n",
        "\n",
        "Portanto, mais uma vez irei começar a procura pelo `n_estimators` com o `learning_rate` instanciado em 0.01, também irei procurar pelo melhor valor para `max_depth`."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "PXjH0q0PQmIF",
        "outputId": "ccd18e45-e4b2-4630-f27c-cc92e4adc512"
      },
      "source": [
        "# Modelo a ser otimizado\n",
        "xgb_smo_Rscaled = XGBClassifier(learning_rate=0.01)\n",
        "\n",
        "# Parâmetros a serem otimizados\n",
        "param_grid = {\n",
        "  'n_estimators':[100, 200, 300, 400],\n",
        "  'max_depth':[1, 2, 3]\n",
        "}\n",
        "\n",
        "# Identificar melhores parâmetros \n",
        "kfold = StratifiedKFold(n_splits=10, shuffle=True)\n",
        "grid_search = GridSearchCV(xgb_smo_Rscaled, param_grid, scoring='recall', n_jobs=-1, cv=kfold)\n",
        "grid_result = grid_search.fit(X_smo_Rscaled, y_train_smo_Rscaled)\n",
        "\n",
        "# Ver resultados\n",
        "print(f'Melhor: {grid_result.best_score_} para {grid_result.best_params_}')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Melhor: 0.9028350515463917 para {'max_depth': 1, 'n_estimators': 100}\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "wN6XL7d0JZI5"
      },
      "source": [
        "Se olharmos para a métrica, temos um valor impressionante de **0.9028**, porém é improvável que esse valor seja mantido quando o modelo for testado.\n",
        "\n",
        "Com os valores para `max_depth` e `n_esimators` obtidos, vou realizar a busca pelos parâmetros `min_child_weight` e `gamma`. Após a busca, será realizada a verificação da melhor taxa de `learning_rate` para o modelo com os parâmetros otimizados."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "kA6BI7_DRsFo",
        "outputId": "95dc57b3-d6b0-4788-9c37-7ae29872ea3b"
      },
      "source": [
        "# Modelo a ser otimizado\n",
        "xgb_smo_Rscaled = XGBClassifier(learning_rate=0.01, n_estimators=100, max_depth=1)\n",
        "\n",
        "# Parâmetros a serem otimizados\n",
        "param_grid = {\n",
        "  'min_child_weight':[1, 3, 5],\n",
        "  'gamma':[0, 1, 5]\n",
        "}\n",
        "\n",
        "# Identificar melhores parâmetros \n",
        "kfold = StratifiedKFold(n_splits=10, shuffle=True)\n",
        "grid_search = GridSearchCV(xgb_smo_Rscaled, param_grid, scoring='recall', n_jobs=-1, cv=kfold)\n",
        "grid_result = grid_search.fit(X_smo_Rscaled, y_train_smo_Rscaled)\n",
        "\n",
        "# Ver resultados\n",
        "print(f'Melhor: {grid_result.best_score_} para {grid_result.best_params_}')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Melhor: 0.9077319587628866 para {'gamma': 0, 'min_child_weight': 1}\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "_BtdGGfKKhtR"
      },
      "source": [
        "Com a otimização dos parâmetros, mais uma vez houve um aumento na métrica, nada muito relevante. Porém, é possível que com uma maior quantidade de dados para treinar o modelo poderia manter uma métrica próxima nos testes.\n",
        "\n",
        "O próximo passo é realizar a busca da melhor taxa para `learning_rate`, uma vez que os parâmetros já foram otimizados."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "9nNHpo4eR-VM",
        "outputId": "b3f99775-c98a-488f-b631-b60e9c97203f"
      },
      "source": [
        "# Modelo a ser otimizado\n",
        "xgb_smo_Rscaled = XGBClassifier(n_estimators=100, max_depth=1, min_child_weight=1, gamma=0)\n",
        "\n",
        "# Parâmetros a serem otimizados\n",
        "param_grid = {\n",
        "  'learning_rate':[0.01, 0.02, 0.04, 0.1]\n",
        "}\n",
        "\n",
        "# Identificar melhores parâmetros \n",
        "kfold = StratifiedKFold(n_splits=10, shuffle=True)\n",
        "grid_search = GridSearchCV(xgb_smo_Rscaled, param_grid, scoring='recall', n_jobs=-1, cv=kfold)\n",
        "grid_result = grid_search.fit(X_smo_Rscaled, y_train_smo_Rscaled)\n",
        "\n",
        "# Ver resultados\n",
        "print(f'Melhor: {grid_result.best_score_} para {grid_result.best_params_}')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Melhor: 0.9077319587628866 para {'learning_rate': 0.01}\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "WZQXXbIsM-tf"
      },
      "source": [
        "#### Avaliando o Desempenho do Modelo XGB \n",
        "\n",
        "Diferente do primeiro modelo de XGBoost que usava os dados balanceados com RUS, esse não teve uma piora com a otimização de `learning_rate`, devido a isso, usarei o valor que foi dado.\n",
        "\n",
        "Pode-se notar que esse modelo teve um desempenho bem inferior quando comparado seja ao modelo de Random Forest ou de XGBoost com dados RUS, tendo uma métrica **recall = 0.8672**, quando **churn = 1**. \n",
        "\n",
        "É notável que este modelo também possui uma grande quantidade de falsos positivos. Para não dizer que há apenas pontos negativos, podemos ressaltar a baixa quantidade de falsos negativos, mas é importante ressaltar que os falsos negativos podem ter se tornado dados falso positivo, dando origem a essa grande quantidade."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 482
        },
        "id": "98LNrYBPSTxd",
        "outputId": "b6025167-1775-4366-a61c-ec369ccca218"
      },
      "source": [
        "# Modelo final\n",
        "xgb_smo_Rscaled = XGBClassifier(learning_rate=0.01, n_estimators=100, max_depth=1, min_child_weight=1, gamma=0)\n",
        "\n",
        "# Treinar Modelo\n",
        "xgb_smo_Rscaled.fit(X_smo_Rscaled, y_train_smo_Rscaled)\n",
        "\n",
        "# Fazer previsões em cima dos dados de teste\n",
        "y_pred_Xgbsmo = xgb_smo_Rscaled.predict(X_test_Rscaled)\n",
        "\n",
        "# Plotar matriz de confusão\n",
        "skplt.metrics.plot_confusion_matrix(y_test, y_pred_Xgbsmo)\n",
        "\n",
        "# Imprimir Relatório de Classificação\n",
        "print(classification_report(y_test, y_pred_Xgbsmo, digits=4))\n",
        "\n",
        "# Imprimir área sob curva\n",
        "print('AUC: {:.4f}\\n'.format(roc_auc_score(y_test, y_pred_Xgbsmo)))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "              precision    recall  f1-score   support\n",
            "\n",
            "           0     0.9198    0.5495    0.6880      1294\n",
            "           1     0.4099    0.8672    0.5567       467\n",
            "\n",
            "    accuracy                         0.6337      1761\n",
            "   macro avg     0.6649    0.7083    0.6223      1761\n",
            "weighted avg     0.7846    0.6337    0.6531      1761\n",
            "\n",
            "AUC: 0.7083\n",
            "\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 2 Axes>"
            ]
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "XbJSJ3GuNJ98"
      },
      "source": [
        "### Otimizando Hiperparâmetros do Modelo SGDClassifier (SGD)\n",
        "\n",
        "Agora irei trabalhar com um modelo que obteve um ótimo desempenho a trabalhar com o balanceamento **SMOTE** e padronização **RobustScaler**.\n",
        "\n",
        "Dessa vez, serão parâmetros novos com os quais ainda não trabalhei e por isso os resultados podem não ser os melhores, porém com uma boa otimização acredito que seja possível obter resultados relevantes.\n",
        "\n",
        "Portanto, irei começar buscando os melhores valores para os parâmetros `loss` e `penalty`.\n",
        "\n",
        "* `loss` - A função de perda a ser usada. O padrão é \"hinge\", o que dá uma SVM (Support Vector Machine) linear.\n",
        "\n",
        "* `penalty` - A penalidade (também conhecida como termo de regularização) a ser usada. O padrão é o 'l2', que é o regularizador padrão para modelos lineares SVM."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "MpWheR4P99A7",
        "outputId": "17008cb1-dcb1-4640-e253-18c70f35e666"
      },
      "source": [
        "# Modelo a ser otimizado\n",
        "sgd_smo_Rscaled = SGDClassifier()\n",
        "\n",
        "# Parâmetros a serem otimizados\n",
        "param_grid = {\n",
        "    'loss':['hinge', 'log'],\n",
        "    'penalty':['l2', 'l1'],\n",
        "\n",
        "}\n",
        "\n",
        "# Identificar melhores parâmetros \n",
        "kfold = StratifiedKFold(n_splits=10, shuffle=True)\n",
        "grid_search = GridSearchCV(sgd_smo_Rscaled, param_grid, scoring='recall', n_jobs=-1, cv=kfold)\n",
        "grid_result = grid_search.fit(X_smo_Rscaled, y_train_smo_Rscaled)\n",
        "\n",
        "# Ver resultados\n",
        "print(f'Melhor: {grid_result.best_score_} para {grid_result.best_params_}')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Melhor: 0.8239690721649484 para {'loss': 'log', 'penalty': 'l2'}\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "meb_hNLCEa2K"
      },
      "source": [
        "Após obter os valores dos parâmetros que estava buscando, irei realizar a busca por outros 2 parâmetros bem importantes, porém irei definir `n_jobs=-1`.\n",
        "\n",
        "Os parâmetros a serem otimizados, são:\n",
        "\n",
        "* `alpha` - Constante que multiplica o termo de regularização. Quanto maior o valor, mais forte é a regularização.\n",
        "\n",
        "* `max_iter` - O número máximo de passos sobre os dados de treinamento (também conhecidos como \"**epochs**\")."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "rS9h3ypYWAqj",
        "outputId": "22254319-590d-4587-c2f7-59ba0f67ddd6"
      },
      "source": [
        "# Modelo a ser otimizado\n",
        "sgd_smo_Rscaled = SGDClassifier(loss='log', penalty='l2')\n",
        "\n",
        "# Parâmetros a serem otimizados\n",
        "param_grid = {\n",
        "    'alpha':[0.0001, 0.001, 0.01, 0.1, 1.0, 10, 100, 1000],\n",
        "    'max_iter':[1000],\n",
        "    'n_jobs':[-1]\n",
        "  \n",
        "}\n",
        "\n",
        "# Identificar melhores parâmetros \n",
        "kfold = StratifiedKFold(n_splits=10, shuffle=True)\n",
        "grid_search = GridSearchCV(sgd_smo_Rscaled, param_grid, scoring='recall', n_jobs=-1, cv=kfold)\n",
        "grid_result = grid_search.fit(X_smo_Rscaled, y_train_smo_Rscaled)\n",
        "\n",
        "# Ver resultados\n",
        "print(f'Melhor: {grid_result.best_score_} para {grid_result.best_params_}')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Melhor: 0.8170103092783505 para {'alpha': 0.001, 'max_iter': 1000, 'n_jobs': -1}\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "DFvx1EoUNiOQ"
      },
      "source": [
        "#### Avaliando o Desempenho do Modelo SGD\n",
        "\n",
        "Agora que os parâmetros foram otimizados, chegou a hora de avaliar o desempenho do Modelo e ver se ele obteve bons resultados. \n",
        "\n",
        "Dessa forma, sempre importante ficar de olho na coluna `recall` do relatório de classificação, também podemos ficar de olho na **AUC**, uma ótima métrica para mensurar o desempenho do Modelo.\n",
        "\n",
        "Aqui é possível ver que o Modelo obteve excelentes resultados, porém outros modelos obtiveram resultados melhores na hora de prever o churn, mesmo que não tenham os valores mais atraentes nas métricas de avaliação."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 482
        },
        "id": "SpIXm4Cj_BcY",
        "outputId": "1a04029e-9367-4fc6-9ce7-a8d2d624433e"
      },
      "source": [
        "# Modelo final\n",
        "sgd_smo_Rscaled = SGDClassifier(loss='log', penalty='l2', alpha=0.001, max_iter=1000, n_jobs=-1)\n",
        "\n",
        "# Treinar Modelo\n",
        "sgd_smo_Rscaled.fit(X_smo_Rscaled, y_train_smo_Rscaled)\n",
        "\n",
        "# Fazer previsões em cima dos dados de teste\n",
        "y_pred_sgdsmo = sgd_smo_Rscaled.predict(X_test_Rscaled)\n",
        "\n",
        "# Plotar matriz de confusão\n",
        "skplt.metrics.plot_confusion_matrix(y_test, y_pred_sgdsmo)\n",
        "\n",
        "# Imprimir Relatório de Classificação\n",
        "print(classification_report(y_test, y_pred_sgdsmo, digits=4))\n",
        "\n",
        "# Imprimir área sob curva\n",
        "print('AUC: {:.4f}\\n'.format(roc_auc_score(y_test, y_pred_sgdsmo)))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "              precision    recall  f1-score   support\n",
            "\n",
            "           0     0.8937    0.7535    0.8176      1294\n",
            "           1     0.5239    0.7516    0.6174       467\n",
            "\n",
            "    accuracy                         0.7530      1761\n",
            "   macro avg     0.7088    0.7525    0.7175      1761\n",
            "weighted avg     0.7956    0.7530    0.7645      1761\n",
            "\n",
            "AUC: 0.7525\n",
            "\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 2 Axes>"
            ]
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "4vDKLQwpNpla"
      },
      "source": [
        "## Método Ensemble para os Modelos do Balanceamento SMOTE\n",
        "\n",
        "Vou usar o método ensemble `VotingClassifier` para combinar o resultado das previsões dos modelos acima. Dessa forma, podemos ter resultados ainda melhores ou mais consistentes.\n",
        "\n",
        "Porém, o modelo com o melhor valor na métrica é o `XGBoost`, o que era de se esperar uma vez que foi o modelo que mais previu corretamente os churn."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 173
        },
        "id": "5h44A8LlcghL",
        "outputId": "6902c91f-bcb7-4199-fc47-9e8cb87c5e33"
      },
      "source": [
        "# Modelos que vão compor o Voting CLassifier\n",
        "rf_ensemble_smo_Rscaled = RandomForestClassifier(bootstrap=False, max_depth=10, max_features=2, min_samples_leaf=3, min_samples_split=6, n_estimators=100)\n",
        "sgd_ensemble_smo_Rscaled = SGDClassifier(loss='log', penalty='l2', alpha=0.001, max_iter=1000, n_jobs=-1) \n",
        "xgb_ensemble_smo_Rscaled = XGBClassifier(learning_rate=0.01, n_estimators=100, max_depth=1, min_child_weight=1, gamma=0)\n",
        "voting_clf_smo = VotingClassifier(estimators=[('RF', rf_ensemble_smo_Rscaled), ('SGD', sgd_ensemble_smo_Rscaled), ('XGB', xgb_ensemble_smo_Rscaled)])\n",
        "\n",
        "\n",
        "# realizar treino dos modelos\n",
        "for model in (rf_ensemble_smo_Rscaled, sgd_ensemble_smo_Rscaled, xgb_ensemble_smo_Rscaled, voting_clf_smo):\n",
        "  model.fit(X_smo_Rscaled, y_train_smo_Rscaled)\n",
        "\n",
        "model = []\n",
        "Recall = []\n",
        "\n",
        "# Mostrar o recall com base nas previsões dos modelos e do Voting Classifier\n",
        "for clf in (rf_ensemble_smo_Rscaled, sgd_ensemble_smo_Rscaled, xgb_ensemble_smo_Rscaled, voting_clf_smo):\n",
        "  y_pred_ensemble = clf.predict(X_test_Rscaled)\n",
        "  model.append(clf.__class__.__name__)\n",
        "  Recall.append(recall_score(y_test, y_pred_ensemble))\n",
        "\n",
        "\n",
        "pd.DataFrame(data=Recall, index=model, columns=[\"Recall\"])"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>Recall</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>RandomForestClassifier</th>\n",
              "      <td>0.706638</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>SGDClassifier</th>\n",
              "      <td>0.758030</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>XGBClassifier</th>\n",
              "      <td>0.869379</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>VotingClassifier</th>\n",
              "      <td>0.755889</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "                          Recall\n",
              "RandomForestClassifier  0.706638\n",
              "SGDClassifier           0.758030\n",
              "XGBClassifier           0.869379\n",
              "VotingClassifier        0.755889"
            ]
          },
          "metadata": {},
          "execution_count": 19
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "VBzUsjUGOI0m"
      },
      "source": [
        "### Otimizando Hiperparâmetros Voting Classifier\n",
        "\n",
        "Mais uma vez, vou realizar a otimização dos parâmetros do `VotingClassifier`, buscando o parâmetro `voting`."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "k6uBNgzpeglE",
        "outputId": "3fbd1b95-95b1-4e7e-9538-8711aad67ab5"
      },
      "source": [
        "# Modelo a ser otimizado\n",
        "voting_clf_smo = VotingClassifier(estimators=[('RF', rf_ensemble_smo_Rscaled), ('SGD', sgd_ensemble_smo_Rscaled), ('XGB', xgb_ensemble_smo_Rscaled)])\n",
        "\n",
        "# Parâmetros a serem otimizados\n",
        "param_grid = {\n",
        "      'voting':('hard', 'soft')\n",
        "}\n",
        "\n",
        "# Identificar melhores parâmetros \n",
        "kfold = StratifiedKFold(n_splits=10, shuffle=True)\n",
        "grid_search = GridSearchCV(voting_clf_smo, param_grid, scoring='recall', n_jobs=-1, cv=kfold)\n",
        "grid_result = grid_search.fit(X_smo_Rscaled, y_train_smo_Rscaled)\n",
        "\n",
        "# Ver resultados\n",
        "print(f'Melhor: {grid_result.best_score_} para {grid_result.best_params_}')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Melhor: 0.8667525773195877 para {'voting': 'hard'}\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "PxqlCE31RUNY"
      },
      "source": [
        "Após otimizar o parâmetro anterior, vou em busca do `n_jobs`."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "98OZtTqVeyBO",
        "outputId": "a02e2756-eefd-413c-92d4-bd7cc93b1ff1"
      },
      "source": [
        "# Modelo a ser otimizado\n",
        "voting_clf_smo = VotingClassifier(estimators=[('RF', rf_ensemble_smo_Rscaled), ('SGD', sgd_ensemble_smo_Rscaled), ('XGB', xgb_ensemble_smo_Rscaled)]\n",
        "                                  , voting='hard')\n",
        "# Parâmetros a serem otimizados\n",
        "param_grid = {\n",
        "      'n_jobs':[1, 3, -1]\n",
        "}\n",
        "\n",
        "# Identificar melhores parâmetros \n",
        "kfold = StratifiedKFold(n_splits=10, shuffle=True)\n",
        "grid_search = GridSearchCV(voting_clf_smo, param_grid, scoring='recall', n_jobs=-1, cv=kfold)\n",
        "grid_result = grid_search.fit(X_smo_Rscaled, y_train_smo_Rscaled)\n",
        "\n",
        "# Ver resultados\n",
        "print(f'Melhor: {grid_result.best_score_} para {grid_result.best_params_}')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Melhor: 0.8711340206185568 para {'n_jobs': -1}\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Pem8XZ8ZOQfs"
      },
      "source": [
        "#### Avaliando o Desempenho do Voting Classifier\n",
        "\n",
        "Para avaliar o `VotingClassifier`, vamos utilizar como referência a coluna `recall` e a `confusion Matrix`. Dessa forma, podemos ter uma noção de quantos churns foram previsto corretamente."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 482
        },
        "id": "qg9Z8E8SfNiO",
        "outputId": "e47ff88d-a9f5-4e45-935a-6d35b8f87abd"
      },
      "source": [
        "# Modelo final\n",
        "voting_clf_smo = VotingClassifier(estimators=[('RF', rf_ensemble_smo_Rscaled), ('SGD', sgd_ensemble_smo_Rscaled), ('XGB', xgb_ensemble_smo_Rscaled)]\n",
        "                                  , voting='hard', n_jobs=-1)\n",
        "\n",
        "# Treinar Modelo\n",
        "voting_clf_smo.fit(X_smo_Rscaled, y_train_smo_Rscaled)\n",
        "\n",
        "# Fazer previsões em cima dos dados de teste\n",
        "y_pred_voting_clf_smo = voting_clf_smo.predict(X_test_Rscaled)\n",
        "\n",
        "# Plotar matriz de confusão\n",
        "skplt.metrics.plot_confusion_matrix(y_test, y_pred_voting_clf_smo)\n",
        "\n",
        "# Imprimir Relatório de Classificação\n",
        "print(classification_report(y_test, y_pred_voting_clf_smo, digits=4))\n",
        "\n",
        "# Imprimir área sob curva\n",
        "print('AUC: {:.4f}\\n'.format(roc_auc_score(y_test, y_pred_voting_clf_smo)))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "              precision    recall  f1-score   support\n",
            "\n",
            "           0     0.9001    0.7032    0.7896      1294\n",
            "           1     0.4880    0.7837    0.6015       467\n",
            "\n",
            "    accuracy                         0.7246      1761\n",
            "   macro avg     0.6940    0.7435    0.6955      1761\n",
            "weighted avg     0.7908    0.7246    0.7397      1761\n",
            "\n",
            "AUC: 0.7435\n",
            "\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 2 Axes>"
            ]
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "BoLo56NpOkjW"
      },
      "source": [
        "## Definindo Modelos para o Balanceamento ADASYN\n",
        "\n",
        "Após realizar todo o trabalho com os modelos que usaram o balanceamento SMOTE vou definir quais serão os 3 modelos usados para o balancemanto **ADASYN** \n",
        "\n",
        "Serão usados os modelos:\n",
        "\n",
        "* **Random Forest (RF)** - com padronização RobustScaler\n",
        "\n",
        "* **XGBoost (XBG)** - com padronização RobustScaler   \n",
        "\n",
        "* **Linear Discriminant Analysis (LDA)** - com padronização RobustScaler\n",
        "\n",
        "Todos os três modelos passarão por um processo de otimização de seus respectivos hiperparâmetros, a fim de obtermos resultados ainda mais consistentes."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 514
        },
        "id": "PfqUIXfbhuK_",
        "outputId": "45f26d60-9392-4e44-bda0-c5ae87487a32"
      },
      "source": [
        "# Imprimir DataSet com os Dados de Recall dos Modelos Escolhidos\n",
        "model = []\n",
        "recall= []\n",
        "\n",
        "model.append('Random Forest adaScaled')\n",
        "recall.append(val_model_balanced(X_ada_scaled, y_train_ada_scaled, rf, quite=True))  \n",
        "model.append('Random Forest adaRscaled')\n",
        "recall.append(val_model_balanced(X_ada_Rscaled, y_train_ada_Rscaled, rf, quite=True))\n",
        "model.append('Random Forest adaQTR')\n",
        "recall.append(val_model_balanced(X_ada_qtr, y_train_ada_qtr, rf, quite=True))\n",
        "model.append('Random Forest adaQTR-N')\n",
        "recall.append(val_model_balanced(X_ada_qtr_n, y_train_ada_qtr_n, rf, quite=True))\n",
        "model.append('Random Forest adaPWR')\n",
        "recall.append(val_model_balanced(X_ada_pwr, y_train_ada_pwr, rf, quite=True))\n",
        "\n",
        "model.append('XGBoost adaScaled')\n",
        "recall.append(val_model_balanced(X_ada_scaled, y_train_ada_scaled, xgb, quite=True))\n",
        "model.append('XGBoost adaRscaled')\n",
        "recall.append(val_model_balanced(X_ada_Rscaled, y_train_ada_Rscaled, xgb, quite=True))\n",
        "model.append('XGBoost adaQTR')\n",
        "recall.append(val_model_balanced(X_ada_qtr, y_train_ada_qtr, xgb, quite=True))\n",
        "model.append('XGBoost adaQTR-N')\n",
        "recall.append(val_model_balanced(X_ada_qtr_n, y_train_ada_qtr_n, xgb, quite=True))\n",
        "model.append('XGBoost adaPWR')\n",
        "recall.append(val_model_balanced(X_ada_pwr, y_train_ada_pwr, xgb, quite=True))\n",
        "\n",
        "model.append('Lda adaScaled')\n",
        "recall.append(val_model_balanced(X_ada_scaled, y_train_ada_scaled, lda, quite=True))  \n",
        "model.append('Lda adaRscaled')\n",
        "recall.append(val_model_balanced(X_ada_Rscaled, y_train_ada_Rscaled, lda, quite=True))\n",
        "model.append('Lda adaQTR')\n",
        "recall.append(val_model_balanced(X_ada_qtr, y_train_ada_qtr, lda, quite=True))\n",
        "model.append('Lda adaQTR-N')\n",
        "recall.append(val_model_balanced(X_ada_qtr_n, y_train_ada_qtr_n, lda, quite=True))\n",
        "model.append('Lda adaPWR')\n",
        "recall.append(val_model_balanced(X_ada_pwr, y_train_ada_pwr, lda, quite=True))\n",
        "\n",
        "pd.DataFrame(data=recall, index=model, columns=['Recall'])"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>Recall</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>Random Forest adaScaled</th>\n",
              "      <td>0.878273</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Random Forest adaRscaled</th>\n",
              "      <td>0.876430</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Random Forest adaQTR</th>\n",
              "      <td>0.880804</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Random Forest adaQTR-N</th>\n",
              "      <td>0.878782</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Random Forest adaPWR</th>\n",
              "      <td>0.872784</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>XGBoost adaScaled</th>\n",
              "      <td>0.853629</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>XGBoost adaRscaled</th>\n",
              "      <td>0.853075</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>XGBoost adaQTR</th>\n",
              "      <td>0.856611</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>XGBoost adaQTR-N</th>\n",
              "      <td>0.858036</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>XGBoost adaPWR</th>\n",
              "      <td>0.853566</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Lda adaScaled</th>\n",
              "      <td>0.784913</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Lda adaRscaled</th>\n",
              "      <td>0.783673</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Lda adaQTR</th>\n",
              "      <td>0.779280</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Lda adaQTR-N</th>\n",
              "      <td>0.796194</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Lda adaPWR</th>\n",
              "      <td>0.772874</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "                            Recall\n",
              "Random Forest adaScaled   0.878273\n",
              "Random Forest adaRscaled  0.876430\n",
              "Random Forest adaQTR      0.880804\n",
              "Random Forest adaQTR-N    0.878782\n",
              "Random Forest adaPWR      0.872784\n",
              "XGBoost adaScaled         0.853629\n",
              "XGBoost adaRscaled        0.853075\n",
              "XGBoost adaQTR            0.856611\n",
              "XGBoost adaQTR-N          0.858036\n",
              "XGBoost adaPWR            0.853566\n",
              "Lda adaScaled             0.784913\n",
              "Lda adaRscaled            0.783673\n",
              "Lda adaQTR                0.779280\n",
              "Lda adaQTR-N              0.796194\n",
              "Lda adaPWR                0.772874"
            ]
          },
          "metadata": {},
          "execution_count": 69
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Lj5lEzAdO4CE"
      },
      "source": [
        "### Otimizando Hiperparâmetros do Modelo Random Forest (RF)\n",
        "\n",
        "Mais uma vez irei usar o Modelo Random Forest para fazer as previsões, sendo um dos modelos com o melhor desempenho.\n",
        "\n",
        "Como ja foi feita uma vez a otimização dos parâmetros, irei apenas repetir esse processo, caso haja alguma diferença no resultado pela busca dos parâmetros."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "4vgKHL5dinSy",
        "outputId": "955f5149-de32-43d6-e79d-ed022dff2a87"
      },
      "source": [
        "# Modelo a ser otimizado\n",
        "rf_ada_qtr = RandomForestClassifier()\n",
        "\n",
        "# Parâmetros a serem otimizados\n",
        "param_grid = {\n",
        "  'bootstrap': [True, False],\n",
        "  'max_depth': [5, 10, 20, 30, 40, 50, None],\n",
        "\n",
        "}\n",
        "\n",
        "# Identificar melhores parâmetros \n",
        "kfold = StratifiedKFold(n_splits=10, shuffle=True)\n",
        "grid_search = GridSearchCV(rf_ada_qtr, param_grid, scoring='recall', n_jobs=-1, cv=kfold)\n",
        "grid_result = grid_search.fit(X_ada_qtr, y_train_ada_qtr)\n",
        "\n",
        "# Ver resultados\n",
        "print(f'Melhor: {grid_result.best_score_} para {grid_result.best_params_}')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Melhor: 0.9147675774520975 para {'bootstrap': False, 'max_depth': 10}\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "pMB6bGJ2Y0f0"
      },
      "source": [
        "Uma vez definido os parâmetros `bootstrap` e `max_depth`, é necessário otimizar os parâmetros `max_features` e `min_samples_leaf.`"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "ymxSHX-5jQjY",
        "outputId": "0455ae8e-f654-4822-9d80-eb7fafcbc64e"
      },
      "source": [
        "# Modelo a ser otimizado\n",
        "rf_ada_qtr = RandomForestClassifier(bootstrap=False, max_depth=10)\n",
        "\n",
        "# Parâmetros a serem otimizados\n",
        "param_grid = {\n",
        "  'max_features': [3, 4, 5, 6],\n",
        "  'min_samples_leaf': [2, 3, 4, 5, 6],\n",
        "\n",
        "}\n",
        "\n",
        "# Identificar melhores parâmetros \n",
        "kfold = StratifiedKFold(n_splits=10, shuffle=True)\n",
        "grid_search = GridSearchCV(rf_ada_qtr, param_grid, scoring='recall', n_jobs=-1, cv=kfold)\n",
        "grid_result = grid_search.fit(X_ada_qtr, y_train_ada_qtr)\n",
        "\n",
        "# Ver resultados\n",
        "print(f'Melhor: {grid_result.best_score_} para {grid_result.best_params_}')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Melhor: 0.9145052076431771 para {'max_features': 3, 'min_samples_leaf': 2}\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "3AFwckzmbWF8"
      },
      "source": [
        "Para podermos avaliar o desempenho do modelo, vamos otimizar os últimos parâmetros `min_samples_split` e `n_estimators`."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "qh_R4Wv_mux9",
        "outputId": "dbc7eb24-70d7-4e2e-df42-86a253db499c"
      },
      "source": [
        "# Modelo a ser otimizado\n",
        "rf_ada_qtr = RandomForestClassifier(bootstrap=False, max_depth=10, max_features=5, min_samples_leaf=2)\n",
        "\n",
        "# Parâmetros a serem otimizados\n",
        "param_grid = {\n",
        "    'min_samples_split':[8, 10, 12, 16],\n",
        "    'n_estimators':[200, 300, 400],\n",
        "\n",
        "}\n",
        "\n",
        "# Identificar melhores parâmetros \n",
        "kfold = StratifiedKFold(n_splits=10, shuffle=True)\n",
        "grid_search = GridSearchCV(rf_ada_qtr, param_grid, scoring='recall', n_jobs=-1, cv=kfold)\n",
        "grid_result = grid_search.fit(X_ada_qtr, y_train_ada_qtr)\n",
        "\n",
        "# Ver resultados\n",
        "print(f'Melhor: {grid_result.best_score_} para {grid_result.best_params_}')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Melhor: 0.9142368748840537 para {'min_samples_split': 8, 'n_estimators': 200}\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "HNCExuIWPCvU"
      },
      "source": [
        "#### Avaliando o Desempenho do Modelo RF\n",
        "\n",
        "Quando comparado ao Modelo que trabalhou com os **dados SMOTE** houve uma piora relativamente pequena, portanto o modelo RF anterior obteve resultados melhores do que esse.\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 486
        },
        "id": "RmnkMj8Kp0FJ",
        "outputId": "e2f30e2c-1e88-4335-d725-d3ac50537d07"
      },
      "source": [
        "# Modelo final\n",
        "rf_ada_qtr = RandomForestClassifier(bootstrap=False, max_depth=10, max_features=5, min_samples_leaf=2, n_estimators=300, min_samples_split=8)\n",
        "\n",
        "# Transformar dados de teste\n",
        "X_test_qtr = qtr.fit_transform(X_test)\n",
        "\n",
        "# Treinar Modelo\n",
        "rf_ada_qtr.fit(X_ada_qtr, y_train_ada_qtr)\n",
        "\n",
        "# Fazer previsões em cima dos dados de teste\n",
        "y_pred_rfAda = rf_ada_qtr.predict(X_test_qtr)\n",
        "\n",
        "# Plotar matriz de confusão\n",
        "skplt.metrics.plot_confusion_matrix(y_test, y_pred_rfAda)\n",
        "\n",
        "# Imprimir Relatório de Classificação\n",
        "print(classification_report(y_test, y_pred_rfAda, digits=4))\n",
        "\n",
        "# Imprimir área sob curva\n",
        "print('AUC: {:.4f}\\n'.format(roc_auc_score(y_test, y_pred_rfAda)))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "              precision    recall  f1-score   support\n",
            "\n",
            "           0     0.8836    0.7921    0.8354      1294\n",
            "           1     0.5524    0.7109    0.6217       467\n",
            "\n",
            "    accuracy                         0.7706      1761\n",
            "   macro avg     0.7180    0.7515    0.7285      1761\n",
            "weighted avg     0.7958    0.7706    0.7787      1761\n",
            "\n",
            "AUC: 0.7515\n",
            "\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 2 Axes>"
            ]
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "A22P-d83PH_q"
      },
      "source": [
        "### Otimizando Hiperparâmetros do Modelo XGBoost (XGB)\n",
        "\n",
        "Esse modelo já teve duas otimizações e o processo é extremamente parecido. Portanto, não irei documentar todos os passos, apenas realizar a otimização para ver qual otimização será retornada para o Modelo."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "pEfXCzJxwosB",
        "outputId": "347e8ff6-ccd6-4e0c-aa12-5393cbcc1aee"
      },
      "source": [
        "# Modelo a ser otimizado\n",
        "xgb_ada_qtr = XGBClassifier(learning_rate=0.01)\n",
        "\n",
        "# Parâmetros a serem otimizados\n",
        "param_grid = {\n",
        "    'n_estimators':[400, 500, 600],\n",
        "    'max_depth':[3, 4, 5, 6]\n",
        "\n",
        "}\n",
        "\n",
        "# Identificar melhores parâmetros \n",
        "kfold = StratifiedKFold(n_splits=10, shuffle=True)\n",
        "grid_search = GridSearchCV(xgb_ada_qtr, param_grid, scoring='recall', n_jobs=-1, cv=kfold)\n",
        "grid_result = grid_search.fit(X_ada_qtr, y_train_ada_qtr)\n",
        "\n",
        "# Ver resultados\n",
        "print(f'Melhor: {grid_result.best_score_} para {grid_result.best_params_}')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Melhor: 0.8964792091802931 para {'max_depth': 3, 'n_estimators': 400}\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "gh4u1kqPxJ2c",
        "outputId": "644c7e77-1aeb-4c39-e37e-1f4a187ece40"
      },
      "source": [
        "# Modelo a ser otimizado\n",
        "xgb_ada_qtr = XGBClassifier(learning_rate=0.01, max_depth=5, n_estimators=400)\n",
        "\n",
        "# Parâmetros a serem otimizados\n",
        "param_grid = {\n",
        "  'gamma':[0, 1, 5],\n",
        "  'min_child_weight':[1, 3, 5]\n",
        "}\n",
        "\n",
        "# Identificar melhores parâmetros \n",
        "kfold = StratifiedKFold(n_splits=10, shuffle=True)\n",
        "grid_search = GridSearchCV(xgb_ada_qtr, param_grid, scoring='recall', n_jobs=-1, cv=kfold)\n",
        "grid_result = grid_search.fit(X_ada_qtr, y_train_ada_qtr)\n",
        "\n",
        "# Ver resultados\n",
        "print(f'Melhor: {grid_result.best_score_} para {grid_result.best_params_}')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Melhor: 0.8918300956722233 para {'gamma': 5, 'min_child_weight': 5}\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "2fg7drOK0sOK",
        "outputId": "5f652b92-483e-4f99-9e4f-107012fb6294"
      },
      "source": [
        "# Modelo a ser otimizado\n",
        "xgb_ada_qtr = XGBClassifier(max_depth=5, n_estimators=400, gamma=1, min_child_weight=1)\n",
        "\n",
        "# Parâmetros a serem otimizados\n",
        "param_grid = {\n",
        "  'learning_rate':[0.01, 0.05, 0.1]\n",
        "}\n",
        "\n",
        "# Identificar melhores parâmetros \n",
        "kfold = StratifiedKFold(n_splits=10, shuffle=True)\n",
        "grid_search = GridSearchCV(xgb_ada_qtr, param_grid, scoring='recall', n_jobs=-1, cv=kfold)\n",
        "grid_result = grid_search.fit(X_ada_qtr, y_train_ada_qtr)\n",
        "\n",
        "# Ver resultados\n",
        "print(f'Melhor: {grid_result.best_score_} para {grid_result.best_params_}')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Melhor: 0.890544086078499 para {'learning_rate': 0.01}\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "vzGYqknBPTog"
      },
      "source": [
        "#### Avaliando o Desempenho do Modelo XGB\n",
        "\n",
        "Quando comparado aos outros dois modelos de XGB, esse obteve os melhores resultados na hora de realizar previsões de **verdadeiros negativos (churn = 0)**. Porém, foi o pior dos modelos na hora de realizar a previsão dos **verdadeiros positivos (churn = 1)**, foi o que menos preveu corretamente a ocorrência do churn, também houve um aumento considerável no número de falsos negativos, enquanto o modelo anterior teve **62 previsões de falsos negativos**, esse teve **mais que o dobro.**\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 482
        },
        "id": "-3hcVJr72Cwz",
        "outputId": "f9846af8-6d1d-416a-cbc8-2a8509131cf1"
      },
      "source": [
        "# Modelo final\n",
        "xgb_ada_qtr = XGBClassifier(learning_rate=0.01, max_depth=5, n_estimators=400, gamma=1, min_child_weight=1)\n",
        "\n",
        "# Treinar Modelo\n",
        "xgb_ada_qtr.fit(X_ada_qtr, y_train_ada_qtr)\n",
        "\n",
        "# Fazer previsões em cima dos dados de teste\n",
        "y_pred_xgbAda = xgb_ada_qtr.predict(X_test_qtr)\n",
        "\n",
        "# Plotar matriz de confusão\n",
        "skplt.metrics.plot_confusion_matrix(y_test, y_pred_xgbAda)\n",
        "\n",
        "# Imprimir Relatório de Classificação\n",
        "print(classification_report(y_test, y_pred_xgbAda, digits=4))\n",
        "\n",
        "# Imprimir área sob curva\n",
        "print('AUC: {:.4f}\\n'.format(roc_auc_score(y_test, y_pred_xgbAda)))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "              precision    recall  f1-score   support\n",
            "\n",
            "           0     0.8734    0.7782    0.8230      1294\n",
            "           1     0.5280    0.6874    0.5972       467\n",
            "\n",
            "    accuracy                         0.7541      1761\n",
            "   macro avg     0.7007    0.7328    0.7101      1761\n",
            "weighted avg     0.7818    0.7541    0.7632      1761\n",
            "\n",
            "AUC: 0.7328\n",
            "\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 2 Axes>"
            ]
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "tqtBWjQsPcpm"
      },
      "source": [
        "### Otimizando Hiperparâmetros do Modelo Linear Discriminant Analysis (LDA)\n",
        "\n",
        "Mais uma vez esse modelo aparecendo aqui, esse modelo obteve um **excelente desempenho** na hora de realizar as previsões, sempre tendo um ótimo número nas métricas de avaliação, como por exemplo, **o recall**.\n",
        "\n",
        "Portanto, já foi um modelo que teve sua **otimização realizada 2x** e que possui poucos parâmetros a serem otimizados e por isso serei breve nesta etapa."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "PqPDB81J2ucB",
        "outputId": "5c3e717b-9099-4430-99b8-15218497e427"
      },
      "source": [
        "# Modelo a ser otimizado\n",
        "lda_ada_qtr = LinearDiscriminantAnalysis()\n",
        "\n",
        "# Parâmetros a serem otimizados\n",
        "param_grid = {\n",
        "    'solver':['lsqr', 'svd', 'eigen'],\n",
        "    'shrinkage':[0, 1, 0.01]\n",
        "      \n",
        "}\n",
        "\n",
        "# Identificar melhores parâmetros \n",
        "kfold = StratifiedKFold(n_splits=10, shuffle=True)\n",
        "grid_search = GridSearchCV(lda_ada_qtr, param_grid, scoring='recall', n_jobs=-1, cv=kfold)\n",
        "grid_result = grid_search.fit(X_ada_qtr, y_train_ada_qtr)\n",
        "\n",
        "# Ver resultados\n",
        "print(f'Melhor: {grid_result.best_score_} para {grid_result.best_params_}')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Melhor: 0.7883152678027191 para {'shrinkage': 1, 'solver': 'lsqr'}\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "2pIOx8ijPoX1"
      },
      "source": [
        "#### Avaliando o Desempenho do Modelo LDA\n",
        "\n",
        "Apesar do excelente desempenho ao treinar com esses dados, pode-se notar que ele obteve resultados melhores quando treinou com dados RUS. \n",
        "\n",
        "Porém, é um **excelente modelo** e com um **desempenho acima da média** quando comparado com alguns outros modelos do projeto, mas ainda **está longe de ser o melhor modelo** e também não é o que eu recomendaria para ser implementado em uma empresa."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 482
        },
        "id": "fmbcjsQI5Ogl",
        "outputId": "f743fd03-9492-409d-9a7a-ccc4c3130008"
      },
      "source": [
        "# Modelo final\n",
        "lda_ada_qtr = LinearDiscriminantAnalysis(solver='lsqr', shrinkage=0.01)\n",
        "\n",
        "# Treinar Modelo\n",
        "lda_ada_qtr.fit(X_ada_qtr, y_train_ada_qtr)\n",
        "\n",
        "# Fazer previsões em cima dos dados de teste\n",
        "y_pred_ldaAda = lda_ada_qtr.predict(X_test_qtr)\n",
        "\n",
        "# Plotar matriz de confusão\n",
        "skplt.metrics.plot_confusion_matrix(y_test, y_pred_ldaAda)\n",
        "\n",
        "# Imprimir Relatório de Classificação\n",
        "print(classification_report(y_test, y_pred_ldaAda, digits=4))\n",
        "\n",
        "# Imprimir área sob curva\n",
        "print('AUC: {:.4f}\\n'.format(roc_auc_score(y_test, y_pred_ldaAda)))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "              precision    recall  f1-score   support\n",
            "\n",
            "           0     0.8973    0.7025    0.7880      1294\n",
            "           1     0.4853    0.7773    0.5975       467\n",
            "\n",
            "    accuracy                         0.7223      1761\n",
            "   macro avg     0.6913    0.7399    0.6928      1761\n",
            "weighted avg     0.7881    0.7223    0.7375      1761\n",
            "\n",
            "AUC: 0.7399\n",
            "\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 2 Axes>"
            ]
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "scVHGMU9PzjG"
      },
      "source": [
        "## Método Ensemble para os Modelos do Balanceamento ADASYN\n",
        "\n",
        "Mais uma vez vou combinar os 3 modelos que usaram os dados ADASYN para treinar, a fim de talvez obter um melhor resultado nas previsões e também resultados mais consistentes ao lidar com **dados novos**."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 175
        },
        "id": "QzunR_E_Lc8Y",
        "outputId": "ae26c46c-6cec-4daa-d2e6-411d19ad3ec5"
      },
      "source": [
        "# Modelos que vão compor o Voting CLassifier\n",
        "rf_ensemble_ada_qtr = RandomForestClassifier(bootstrap=False, max_depth=10, max_features=5, min_samples_leaf=2, n_estimators=300, min_samples_split=8) \n",
        "xgb_ensemble_ada_qtr = XGBClassifier(learning_rate=0.01, max_depth=5, n_estimators=400, gamma=1, min_child_weight=1) \n",
        "lda_ensemble_ada_qtr = LinearDiscriminantAnalysis(solver='lsqr', shrinkage=0.01)\n",
        "voting_clf_ada = VotingClassifier(estimators=[('RF', rf_ensemble_ada_qtr), ('XGB', xgb_ensemble_ada_qtr), ('LDA', lda_ensemble_ada_qtr)])\n",
        "\n",
        "\n",
        "# realizar treino dos modelos\n",
        "for model in (rf_ensemble_ada_qtr, xgb_ensemble_ada_qtr, lda_ensemble_ada_qtr, voting_clf_ada):\n",
        "  model.fit(X_ada_qtr, y_train_ada_qtr)\n",
        "\n",
        "model = []\n",
        "Recall = []\n",
        "\n",
        "# Mostrar o recall com base nas previsões dos modelos e do Voting Classifier\n",
        "for clf in (rf_ensemble_ada_qtr, xgb_ensemble_ada_qtr, lda_ensemble_ada_qtr, voting_clf_ada):\n",
        "  y_pred_ensemble = clf.predict(X_test_qtr)\n",
        "  model.append(clf.__class__.__name__)\n",
        "  Recall.append(recall_score(y_test, y_pred_ensemble))\n",
        "\n",
        "\n",
        "pd.DataFrame(data=Recall, index=model, columns=[\"Recall\"])"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>Recall</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>RandomForestClassifier</th>\n",
              "      <td>0.710921</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>XGBClassifier</th>\n",
              "      <td>0.704497</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>LinearDiscriminantAnalysis</th>\n",
              "      <td>0.811563</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>VotingClassifier</th>\n",
              "      <td>0.740899</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "                              Recall\n",
              "RandomForestClassifier      0.710921\n",
              "XGBClassifier               0.704497\n",
              "LinearDiscriminantAnalysis  0.811563\n",
              "VotingClassifier            0.740899"
            ]
          },
          "metadata": {},
          "execution_count": 23
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "I_pVP3rCQCMp"
      },
      "source": [
        "### Otimizando Hiperparâmetros do Voting Classifier\n",
        "\n",
        "Como das últimas vezes, os parâmetros otimizados são os mesmos. Dessa forma, a otimização é bem parecida com as anteriores."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "zKPymSOWP4Qm",
        "outputId": "7a92bb10-3289-4bb2-f22b-4ce4ddf4b3ef"
      },
      "source": [
        "# Modelo a ser otimizado\n",
        "voting_clf_ada = VotingClassifier(estimators=[('RF', rf_ensemble_ada_qtr), ('XGB', xgb_ensemble_ada_qtr), ('LDA', lda_ensemble_ada_qtr)], voting='hard')\n",
        "\n",
        "# Parâmetros a serem otimizados\n",
        "param_grid = {\n",
        "      'voting':['soft', 'hard'],\n",
        "      'n_jobs':[1, 2, -1]\n",
        "}\n",
        "\n",
        "# Identificar melhores parâmetros \n",
        "kfold = StratifiedKFold(n_splits=10, shuffle=True)\n",
        "grid_search = GridSearchCV(voting_clf_ada, param_grid, scoring='recall', n_jobs=-1, cv=kfold)\n",
        "grid_result = grid_search.fit(X_ada_qtr, y_train_ada_qtr)\n",
        "\n",
        "# Ver resultados\n",
        "print(f'Melhor: {grid_result.best_score_} para {grid_result.best_params_}')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Melhor: 0.8857659856651052 para {'n_jobs': -1, 'voting': 'hard'}\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "L28W7pjyQIyG"
      },
      "source": [
        "#### Avaliando o Desempenho do Voting Classifier\n",
        "\n",
        "Os modelos de Voting Classifier possuem resultados bem parecidos, até mesmo porque são constituídos quase que dos mesmos modelos, diferenciando apenas o segundo Voting Classifier que **não possui o modelo LDA** e sim o **SGD**, devido a isso, as previsões são parecidas e cometem os mesmo erros.\n",
        "\n",
        "Porém, esse modelo foi o que menos fez previsões de **falsos negativos e de falsos positivos** também, isso pode ser benéfico para o consumidor que não sofrerá com as ligações indesejadas, mas provavelmente não é objetivo das empresas."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 486
        },
        "id": "GV_dNZk-Q9t-",
        "outputId": "dffe496a-7afe-4773-e623-c28ac2c2a0e8"
      },
      "source": [
        "# Modelo final\n",
        "voting_clf_ada = VotingClassifier(estimators=[('RF', rf_ensemble_ada_qtr), ('XGB', xgb_ensemble_ada_qtr), ('LDA', lda_ensemble_ada_qtr)], voting='hard', n_jobs=-1)\n",
        "\n",
        "# Treinar Modelo\n",
        "voting_clf_ada.fit(X_ada_qtr, y_train_ada_qtr)\n",
        "\n",
        "# Fazer previsões em cima dos dados de teste\n",
        "y_pred_voting_clf_ada = voting_clf_ada.predict(X_test_qtr)\n",
        "\n",
        "# Plotar matriz de confusão\n",
        "skplt.metrics.plot_confusion_matrix(y_test, y_pred_voting_clf_ada)\n",
        "\n",
        "# Imprimir Relatório de Classificação\n",
        "print(classification_report(y_test, y_pred_voting_clf_ada, digits=4))\n",
        "\n",
        "# Imprimir área sob curva\n",
        "print('AUC: {:.4f}\\n'.format(roc_auc_score(y_test, y_pred_voting_clf_ada)))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "              precision    recall  f1-score   support\n",
            "\n",
            "           0     0.8939    0.7813    0.8338      1294\n",
            "           1     0.5508    0.7430    0.6326       467\n",
            "\n",
            "    accuracy                         0.7712      1761\n",
            "   macro avg     0.7223    0.7622    0.7332      1761\n",
            "weighted avg     0.8029    0.7712    0.7805      1761\n",
            "\n",
            "AUC: 0.7622\n",
            "\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAT4AAAEWCAYAAAD/x/trAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAgAElEQVR4nO3de1xUdf748dcBvIACAgl420zNdNHU1JRfqIkNqCAgytpu2YrbqkWSmqbVmqZiZpaUZS1ZeFnLVlPwVqKY4BUhJcuyNvOCF2Dl7o3LcH5/sMw3UmFmnIEZ5v3scR4P5syZ83kPyfvx+ZxzPp+3oqqqihBC2BC7hg5ACCHqmyQ+IYTNkcQnhLA5kviEEDZHEp8QwuZI4hNC2BxJfI3MzZs3mTJlCn379iU6Otro82zdupWJEyeaMLKG8fTTT7Nly5aGDkNYGEWe42sY27ZtIz4+njNnztCiRQu6devGlClT6Nev312dNyEhgX/9619s2LABBwcHE0VrOmlpaTz11FM89thjvP/++7r9p06dIjQ0lIcffph169bVeZ4VK1Zw7tw5li1bZs5wRSNleX8ZNiA+Pp64uDhee+01/Pz8aNKkCfv37yc5OfmuE9+lS5fo2LGjRSa9au7u7mRmZlJQUICbmxsAW7ZsoWPHjiZrQ1VVVFXFzk4GNeJW8q+inpWUlPDuu+/y6quvEhAQgJOTE02aNMHf35/Zs2cDUFZWRkxMDH5+fvj5+RETE0NZWRlQ1WMaPHgwn3zyCb6+vvj5+fHFF18A8O6777Jy5Uq+/PJL+vTpw8aNG1mxYgUzZ87UtX/hwgUeeOABKioqANi8eTPDhg2jT58++Pv7s3XrVt3+P//5z7rPHTt2jDFjxtC3b1/GjBnDsWPHdO+NHz+e2NhYHn/8cfr06cPEiRPJz8+/4++gSZMmDBs2jJ07dwKg1WrZuXMno0aNqnHcokWLGDJkCA899BDh4eFkZGQAkJqayj//+U/d9wwJCdHFsXz5ch5//HF69epFVlYW48ePZ+PGjQDMmzePqVOn6s7/5ptv8te//hUZ9NgeSXz17Pjx45SWlqLRaO54zAcffMC3335LYmIiW7du5bvvvmPlypW6969cuUJJSQmpqanExMSwYMECioqKiI6OZvLkyYwYMYLjx48TERFRayzXr19n0aJFfPTRRxw/fpwNGzbQvXv3W44rLCxk8uTJjB8/nrS0NCIjI5k8eTIFBQW6Y7Zv387rr7/O4cOHKS8v55NPPqm17bCwMBISEgA4cOAAXbt2xcvLq8YxPXv2JCEhgaNHjxIcHMzzzz9PaWkpgwcPrvE9q5M1QGJiIgsXLuTYsWO0bdu2xvnmzJnDzz//zObNm8nIyGDTpk288cYbKIpSa6yi8ZHEV88KCwtxc3OrdSi6bds2oqKi8PDwwN3dnaioqBp/3A4ODkRFRdGkSROGDBmCk5MTZ86cMSoeOzs7/vOf/3Dz5k08PT25//77bzlm37593HvvvYSFheHg4EBwcDCdOnXi66+/1h0THh7OfffdR/PmzRk+fDg//vhjre0+9NBDFBUV8euvv5KQkEBoaOgtx4SGhup+VxMnTqSsrKzO7zl69Gjuv/9+HBwcaNKkSY33HB0dWbp0KUuWLGHWrFnMnTsXb2/vWs8nGidJfPWsVatWFBQU6Iaat5Obm1ujt9K2bVtyc3NrnOO3idPR0ZHr168bHIuTkxPLly9nw4YN+Pn5MWnSJE6fPl1nPNUx5eTk6F63bt3a4HhCQkJYv349aWlpt+0Bf/zxx4wYMYK+ffvSr18/SkpKavQyb6dNmza1vt+rVy/at2+PqqqMGDGizhhF4ySJr5716dOHpk2bsmfPnjse4+npyaVLl3SvL1++jKenp1HtOTo6cvPmTd3rK1eu1Hh/0KBBxMfHc+DAATp16sTcuXPrjKc6pt8PTQ0VGhrKp59+ypAhQ3B0dKzxXkZGBqtWrSI2Npb09HQyMjJwdnbWXY+70/C0rmHr+vXrKS8vx9PTk1WrVt1V/MJ6SeKrZ87OzkRHR7NgwQL27NnDjRs3KC8vJyUlhaVLlwIQFBTEBx98QH5+Pvn5+bz//vu3XPjXV/fu3UlPT+fSpUuUlJTwz3/+U/felStX2LNnD9evX6dp06Y4OTnd9i7okCFDOHv2LNu2baOiooKdO3fyyy+/8OijjxoVU7UOHTqwbt06pk2bdst7165dw97eHnd3dyoqKnjvvfe4evWq7n0PDw8uXrxIZWWl3u2dOXOG2NhY3nzzTZYuXcqqVavqHJKLxkkSXwOYOHEic+bMYeXKlfj6+vLoo4+yfv16HnvsMQCeffZZevToQUhICCEhIfj4+PDss88a1dYjjzzCyJEjCQkJITw8nKFDh+req6ysZPXq1QwaNIiHH36Y9PR05s+ff8s53Nzc+PDDD4mPj2fAgAGsWrWKDz/8EHd3d6Ni+q1+/frdtufo5+fHoEGDCAwMxN/fn2bNmtUYxg4fPhyAAQMGMHr06DrbqaioYNasWfz973+nW7dudOzYkenTp/Piiy/q7pgL2yEPMAshbI70+IQQNkcSnxDC5kjiE0LYHEl8QgibY1Ez2a8UXOXc5TvP8RSW549d2jV0CMIATewVHOzvrr+TdPAHPNxa6nVs3z/+odb3X3rpJfbt24eHhwfbt28HqmY3TZ8+nYsXL9KuXTtiY2NxdXVFVVViYmJISUmhefPmLFmyBB8fH6BqkYsPPvgAgGeeeabOO/0WdVf3mx/O4/fE0oYOQxggfduShg5BGKCzpyOOTe3v6hyG/J3eOP5ere+np6fj5OTE7NmzdYlv6dKltGrVikmTJhEXF0dRURGzZs0iJSWFdevW8dFHH/Htt98SExPDxo0bKSwsZMyYMXzxxRcoikJ4eDibN2/G1dX1ju3KUFcIYTjFTr+tDv37978lQSUnJxMWFgZULWZRPcuper+iKPTu3Zvi4mJyc3M5cOAAjzzyCK1atcLV1ZVHHnmE/fv319quRQ11hRDWQAE7/XqN+fn5PP3007rX48aNY9y4cbV+Ji8vTzdFs3Xr1uTl5QGQk5NTY1EJb29vcnJybtnv5eVVYx757UjiE0IYTs+lvNzd3dm8efNdNKOYZdkwGeoKIQyjYLKh7u14eHjoViPKzc3VTY308vIiOztbd1x2djZeXl637M/JyalzAQ1JfEIIwymKfpsR/P39dYvUJiQkMGzYsBr7VVUlMzMTZ2dnPD098fPz48CBAxQVFVFUVMSBAwfw8/OrtQ0Z6gohDKQY3Zv7vRkzZnD06FEKCgoYPHgwU6dOZdKkSUybNo1NmzbRtm1bYmNjgapVglJSUtBoNDg6OrJ48WKgan3KZ599lrFjxwIQFRVFq1atav8G8jiLuBvyOIt1McnjLD9ewG9i7Y+pVLtx2DL/fUiPTwhhGEX/u7qWShKfEMJwJhrqNhRJfEIIw1l5ZTpJfEIIA5nu5kZDkcQnhDCcJD4hhE1RFLCXmxtCCFsj1/iEEDZHhrpCCJuiID0+IYStkbu6QghbJD0+IYRtkSlrQghbU70enxWTxCeEMJwMdYUQtsX6b25Yd/RCiIZhoqXn16xZQ3BwMEFBQaxevRqoqqsbGRlJQEAAkZGRFBUVAaCqKosWLUKj0TBq1ChOnjxpdPiS+IQQhqlej0+frRY///wzGzduZOPGjSQmJrJv3z7OnTtHXFwcvr6+JCUl4evrS1xcHACpqamcPXuWpKQkFi5cyPz5843+CpL4hBCGM0HNjdOnT/Pggw/i6OiIg4MD/fv3JykpyeC6usaQxCeEMJyeQ938/HzCw8N12+eff647RdeuXfnmm28oKCjgxo0bpKamkp2dbXBdXWPIzQ0hhOFMUFe3c+fOPP300/ztb3/D0dGRbt26YWdXsy8mdXWFEBZBQdElpLq2ukRERLB582bWr1+Pq6srHTt2NLiurjEk8QkhDKNgssRXPYy9dOkSSUlJjBo1yuC6usaQoa4QwmCKnWmGn1OnTqWwsBAHBwfmzZuHi4uLwXV1jSGJTwhhMFNdd/v0009v2efm5saaNWtu2+a8efNM0q4kPiGEwcxxw6E+SeITQhjEXHda65MkPiGE4aw770niE0IYTnp8Qgib8/sHja2NJD4hhEEURXp8QghbZN15TxKfEMJQcldXCGGDJPEJIWyLYropaw1FEp8QwmDS4xNC2BQFSXxCCJsjNzeEELZGnuMTQtgk6857kviEEIaz9ilr1h29EKLembLmxurVqwkKCiI4OJgZM2ZQWlpKVlYWERERaDQapk2bRllZGQBlZWVMmzYNjUZDREQEFy5cMPo7SOITQhhGMWCrRU5ODmvXruWLL75g+/btaLVaduzYwbJly5gwYQK7d+/GxcWFTZs2AbBx40ZcXFzYvXs3EyZMYNmyZUZ/BUl8Rvhw3hOcS36djI0v6/a5uTix/YPn+C7xVbZ/8BytnB0B6NrRi31rXqAwbTnTxg+r8zzC/LIvXWDin0YS6t+PsGH9+dfHKwE4dfIET4QMZWzg/2PcyMF8dzwDgL27thOuGajbf+zooYYM3yKYqsen1Wq5efMmFRUV3Lx5k9atW3PkyBECAwMBGD16NMnJyQDs3buX0aNHAxAYGMjhw4dRVdWo+M2a+FJTUwkMDESj0RAXF2fOpurVum1HCI16v8a+mZEa9h39iZ6hC9h39CdmRgYAUFB0jRfe2Ejs2r16nUeYn729AzPnLiZxbwbrE/eyYU0cp38+xdsxc5ky/SU27TpE1MxXeHvxXAAG+j3KF0mH2bTrEAveWsm8F59r4G/Q8PRNfLUVFPfy8mLixIkMHToUPz8/WrZsiY+PDy4uLjg4VN1++G3R8JycHNq0aQOAg4MDzs7OFBQUGBW/2W5uaLVaFixYQHx8PF5eXowdOxZ/f3+6dOliribrzcFjp/lDG/ca+4IffZDAv78DwL+2pbHro+f5x7uJ/LfgKv8tuMrwQT30Oo8wv9Ze3rT28gagRUtn7uvyADnZl1AUhWslJQBcLS6mtVfVH5lTi5a6z964fs3qH+UwBX1/B7UVFC8qKiI5OZnk5GScnZ15/vnn2b9/vynDvCOzJb4TJ05w77330qFDBwCCgoJITk5uFInvdjw9nMm+UgxA9pViPD2cGzgioY+LWec4dfIED/bpx+z5S5j85GiWLXoFtbKSdQl7dMclf7mV2Dfmk3/lCu+v2diAEVsAE83VPXToEO3bt9cVDA8ICODYsWMUFxdTUVGBg4NDjaLhXl5eXL58GW9vbyoqKigpKcHNzc2ots021M3JycHb21v32svLS9dltQVGXnoQ9ej6tatMn/wks+cvoaWzC5+v+5gX5y1hz9FTzJq3hFdnRemOHTYihG37jvHOqk95b9miBoy64Znqrm7btm359ttvuXHjBqqqcvjwYbp06cKAAQPYtWsXAFu2bMHf3x+oKii+ZcsWAHbt2sXAgQON7n3LzQ0Tyc0rwfseFwC873Hhv/klDRyRqE15eTnTJz1JUNifeGxEKABbN33KYyNCAAgMHs33md/c8rl+A/24cP4sBflX6jVeS2OKxNerVy8CAwMZPXo0o0aNorKyknHjxjFr1izi4+PRaDQUFhYSEREBwNixYyksLESj0RAfH8/MmTONjt9sQ10vLy+ys7N1r3NycnRd1sZoR8p3PDlqAMvid/PkqAFs33eioUMSd6CqKvNmRdHp/gf466Spuv2tvbzJOHKA/r6DSDuYwh/u6wzA+TOn6dCxE4qi8MN3mZSXltLKzaOhwrcIprrMGR0dTXR0dI19HTp00D3C8lvNmjXj3XffNUm7Zkt8PXv25OzZs2RlZeHl5cWOHTt46623zNVcvVrz+gQG9b2fe1q15JevFrLww50si9/Nv96YyF/DfDl/OZ8nX/wEAC8PZw6ufxHnFs2pVFWee+JR+oyJoeTazdueZ03C4Qb+do3f8fTDbPviM+7v5sPYwP8HQPTsecx/YwVL5s9GW1FBs2bNmbek6o9s95eJbPviMxwcmtCseXPeXLna5m9wWPv3V1RjH4TRQ0pKCosXL0ar1TJmzBieeeaZWo//5ofz+D2x1FzhCDNI37akoUMQBujs6YhjU/u7Osf3F4oYs+KIXsf+9EbgXbVlLmadqztkyBCGDBliziaEEA3Ayjt8skiBEMJwdrL0vBDCpijS4xNC2BhFVmAWQtgiK897kviEEIaz9oVIJfEJIQwj1/iEELZGyksKIWySlec9SXxCCMNJj08IYXOsPO9J4hNCGEZRFJm5IYSwPTLUFULYHFPkvV9//ZXp06frXmdlZREdHU1YWBjTp0/n4sWLtGvXjtjYWFxdXVFVlZiYGFJSUmjevDlLlizBx8fHqLat+ylEIUSDMMUKzJ06dSIxMZHExEQ2b96Mo6OjriKjr68vSUlJ+Pr66io0pqamcvbsWZKSkli4cCHz5883On5JfEIIw/zvAWZ9Nn0dPnyYDh060K5dO5KTkwkLCwMgLCyMPXuqij5V71cUhd69e1NcXExubq5RX0GGukIIgyjovyxVfn4+Tz/9tO71uHHjGDdu3C3H7dixg+DgYADy8vLw9PQEoHXr1uTl5QG3FjCrrrlbfawhJPEJIQxmirq61crKyti7dy8vvPDCbdsxx40UGeoKIQxkmvKS1VJTU/Hx8eGee+4BwMPDQzeEzc3N1dXd/X0Bs9/W3DWUJD4hhGFMfI1vx44dBAUF6V77+/uTkJAAQEJCAsOGDauxX1VVMjMzcXZ2NmqYC5L4hBAGql6kwBQ9vuvXr3Po0CECAgJ0+yZNmsTBgwcJCAjg0KFDTJo0Caiq4dOhQwc0Gg1z585l3rx5Rn8HucYnhDCYqS67OTk5kZaWVmOfm5sba9asuU2byl0lu9+SxCeEMJhMWRNC2BRFATuZsiaEsDVWnvck8QkhDCeLFAghbI6VX+K7c+JbuHBhrVn9H//4h1kCEkJYtka9Hl+PHj3qMw4hhBVRaKSJb/To0TVe37hxA0dHR7MHJISwfFbe4at75sbx48cZOXIkI0aMAODUqVN3tQ6WEMK6mXLmRkOpM/EtXryYjz/+mFatWgHQrVs3MjIyzB6YEMJymXo9vvqm113dNm3a1HhtZydTfIWwZY3+AeY2bdpw7NgxFEWhvLyctWvX0rlz5/qITQhhgRTF+qes1dl1mz9/PuvXrycnJ4dBgwbx448/8uqrr9ZHbEIIC9Xoh7ru7u689dZb9RGLEMIqKFY/1K2zx5eVlcWUKVMYOHAgvr6+PPPMM2RlZdVHbEIIC6XouVmqOhPfCy+8wPDhwzlw4AD79+9n+PDhzJgxoz5iE0JYIJt4nOXGjRuEhYXh4OCAg4MDoaGhlJaW1kdsQghLpFQ9wKzPVpfi4mKio6MZPnw4I0aM4Pjx4xQWFhIZGUlAQACRkZEUFRUBoKoqixYtQqPRMGrUKE6ePGn0V7hj4issLKSwsJDBgwcTFxfHhQsXuHjxIh999BFDhgwxukEhhPWzs1P02uoSExPDoEGD+Oqrr0hMTKRz5871UlD8jjc3wsPDURQFVVUB2LBhg+49RVFuWwpOCNH4VQ9171ZJSQnp6eksWbIEgKZNm9K0aVOSk5NZt24dUFVQfPz48cyaNeuOBcVNWld37969Rn4dIURjp+9jfLUVFL9w4QLu7u689NJLnDp1Ch8fH1555RXLKSj+888/88svv1BWVqbbFxYWZnBjQohGQDFNQfGKigp++OEH5s6dS69evVi0aJFuWKtrykw3SepMfO+99x5paWmcPn2aIUOGkJqaSt++fSXxCWGjTPWoire3N97e3vTq1QuA4cOHExcXpyso7unp2XAFxXft2sWaNWu45557eP3110lMTKSkpMSoxoQQjYGCvZ1+W21at26Nt7c3v/76KwCHDx+mc+fO9VJQvM4eX7NmzbCzs8PBwYGrV6/i4eHB5cuXjWpMCNE4mGr4OXfuXGbOnEl5eTkdOnTg9ddfp7KykmnTprFp0ybatm1LbGwsUFVQPCUlBY1Gg6OjI4sXLza63ToTX48ePSguLiYiIoLw8HCcnJzo06eP0Q0KIaxb1V1d05yre/fut70G2OAFxauflfnzn//MoEGDuHr1Kt26dTNJ40IIK9SY6+rW9lT0yZMn8fHxMUtAQgjLZ+V5786Jr/qhwttRFIW1a9eaPJg+3f9AQfp7Jj+vMJ9vzhQ0dAjCANrK5iY5jyXPw9XHHRNf9ZPTQgjxWwpg31gTnxBC3JaeCxBYMkl8QgiDKEjiE0LYHMtea08fdc7cUFWVxMRE3nuv6qbDpUuXOHHihNkDE0JYLlOtx9dQ9Co2lJmZyY4dOwBo0aIFr732mtkDE0JYpuoHmK252FCdie/EiRPMmzePZs2aAeDq6kp5ebnZAxNCWCgFHBRFr81S1XmNz8HBAa1WqxvT5+fnS0FxIWyYKaesNZQ6E9/48eOJiooiLy+P5cuX89VXXzFt2rT6iE0IYaEa7ZS1aiEhIfj4+HDkyBFUVWXlypV07ty5PmITQlgoK897dSe+S5cu4ejoyNChQ2vsa9u2rVkDE0JYJpt4jm/y5Mm6n0tLS7lw4QL33Xef7i6vEMLGKNS5yKilqzPxbdu2rcbrkydP8umnn5otICGEZTNlj8/f358WLVpgZ2eHvb09mzdvprCwkOnTp3Px4kXatWtHbGwsrq6uqKpKTEwMKSkpNG/enCVLlhi9SpTBt2d9fHzkAWYhbJyi53/6WLNmDYmJiboFSRu0rm61+Ph43c+VlZX88MMPRq9zL4RoHMw50m3QurrVrl27pvvZ3t6eIUOGEBgYaHBDQojGwZChbm11dav97W9/Q1EU3XsNXldXq9Vy7do1Zs+ebfCJhRCNlAG1bmurqwvw2Wef4eXlRV5eHpGRkXTq1Ol3TZlnQYQ7XuOrqKjA3t6eY8eOmbxRIYR1s7fTb6tLdV1cDw8PNBoNJ06c0NXVBeq/rm5ERAQA3bp1Y8qUKSQkJJCUlKTbhBC2qWqoq+i11eb69etcvXpV9/PBgwe5//77LaOubllZGW5ubqSlpdXYHxAQYFSDQgjrZ4qbG3l5eURFRQFVl9WCg4MZPHgwPXv2NHtdXUVVVfV2bwwePJjIyEhUVUVRFH57mKIoREZGGt3onVSqUKY1+WmFGUmxIevSq4MzLZvf3frDOVdL+fe3l/Q6duoj991VW+Zyx99AZWVljTu6QggB/xvq6vmMnqW6Y+Jr3bo1zz33XH3GIoSwEo12kYI7jICFEDZOARwa61zd1atX12MYQghr0mh7fK1atarPOIQQ1kKxgYVIhRDit2xi6XkhhPg9a6+6I4lPCGGgumdlWDpJfEIIg1RPWbNmkviEEAaz7rQniU8IYQQr7/BJ4hNCGEjBLGvk1SdJfEIIg1TN1bVukviEEAaTmxtCCJtj7UNda++xCiHqWfVQV59NH1qtlrCwMCZPngxAVlYWERERaDQapk2bRllZGVC1KPK0adPQaDRERERw4cIFo7+DJD4hhIEUXRGgujZ9rF27ls6dO+teL1u2jAkTJrB7925cXFzYtGkTABs3bsTFxYXdu3czYcIEli1bZvQ3kMQnhDCYoudWl+zsbPbt28fYsWOBquXwjhw5oithO3r0aJKTkwHYu3cvo0ePBiAwMJDDhw8bvXyeXOMTQhhEAez17M3VVVd38eLFzJo1S7fae0FBAS4uLjg4VKWm6tq5UFVXt02bNgA4ODjg7OxMQUGBrgqbISTxCSEMo+j/AHNtdXW//vpr3N3d6dGjxy3FzMxNEp8QwmCKCSatHTt2jL1795KamkppaSlXr14lJiaG4uJiKioqcHBwqFE718vLi8uXL+Pt7U1FRQUlJSW4ubkZ1bZc4xNCGKR6PT59ttq88MILpKamsnfvXt5++20GDhzIW2+9xYABA9i1axcAW7Zswd/fH6iqq7tlyxYAdu3axcCBA41+rEYSnxDCYHYoem3GmDVrFvHx8Wg0GgoLC4mIiABg7NixFBYWotFoiI+PZ+bMmUbHf8e6ug1B6upaH6mra11MUVe36EY5R84U6nVs4B9b31Vb5iLX+IQQBpH1+IQQNsnKq0tK4hNCGEoxyV3dhiSJTwhhGAOe47NUclfXBCY/PZE/tPWkb+8eun0vzZ5Frx7d6N/nQf40djSFhf93MfjNN17Hp1sXHvR5gN1JuxoiZJtWWnqTv495jL+OGsSTI335+J3Xa7wfu3AOmt4ddK/fXfwyE0IGMyFkMI8H9Gd43471HLFlqZqOpt9/lspsie+ll17C19eX4OBgczVhMcb/dQKJ27+qsW/YYxq+yfye9OMnuP/+rrz5RtUf148//MDGzzdw7NuTbN3+Fc9PfRatVm5l16emTZvxztoE1mzbz+rEVI7sT+b7zHQATn13nJKimncso19ezOqtqazemsrYJ//O4IDG/2+6LnaKfpulMlviCw8PZ9WqVeY6vUXxGzT4lvmCj2kCdPMNHx4wkIv/W0Jn+7ZEIsY9TrNmzeh433107tyF9KNH6z1mW6YoCk4tWgJQUVGOtqICRVHQarW8v3Qez7w4/46f3bPjCzTBY+opUstlpyh6bZbKbImvf//+uLq6muv0VmXt6k8IHD4CgIsXL9K+/f8No9q1a8+lSxcbKjSbpdVqmRAymFG+D9DvkUfx6dWPL/71EX7+w7nH0/u2n8m+mMXlC+d5aODgeo7Wsui7Movlpj25xmd2b7weg72DA4//5YmGDkX8hr29Pau3prI59Xt+PHGMzPRDfP1lImPGT7rjZ/bs2MyjgSHY29vXY6SWSXp84o7WrVnNzh3bWb12vW5OYbt27bhwIUt3zMWLF2jbtl1DhWjznF1ceWiAH8eO7Ofi+TM8runL2KG9uHnjOuMe61vj2OQdm3ksOLyBIrUs0uMTt5W06yvefmspm7ZsxcnJSbc/KDiEjZ9voLS0lLNnzvDLL/+h/8MPN2Cktqcg/wolxUUAlN68QfrBfTzQozdbD51i09ffsunrb2nu6MTne77Rfebc6Z8pKS6kRx/5fwVYfeaT5/hM4Kkn/8z+lH1cuXKFzh3bM/fV13hz6euUlpYSPFwDVF4Nn60AAArjSURBVN3gWLHyQ/7o48OYiD/R58E/4uDgQOy778vQqZ7l5eYQM/tZKiu1VFZW4j8ijEeGBtb6mT07NjNsZLjVF9kxBUWx/ilrZlukYMaMGRw9epSCggI8PDyYOnWqbpWFO5FFCqyPLFJgXUyxSMG10gp+uHhNr2P7d7LMG5xm6/G9/fbb5jq1EKKhWXeHT4a6QghDWfasDH1I4hNCGMzKL/FJ4hNCGM4Uea+0tJQnnniCsrIytFotgYGBREdHk5WVxYwZMygsLMTHx4elS5fStGlTysrKePHFFzl58iStWrVi+fLltG/f3qi25XEWIYRBqmpu3H1B8aZNm7JmzRq2bt1KQkIC+/fvJzMzUwqKCyEskymKDSmKQosWLQCoqKig4n9zpqWguBDCIuk71K2roLhWqyU8PJzz58/zl7/8hQ4dOkhBcSGEBTJgVkZtBcWhas50YmIixcXFREVF8euvv5omxjrIUFcIYTBTL0Tq4uLCgAEDyMzM1BUUB25bUByQguJCiPpnimt8+fn5FBcXA3Dz5k0OHTpE586d66WguAx1hRAGqbqre/fnyc3NZc6cOWi1WlRVZfjw4QwdOpQuXbowffp0YmNj6d69e42C4rNmzUKj0eDq6sry5cuN/w5SUFzcDZmra11MMVf3RpmWX/97U69jfdq1uKu2zEV6fEIIg8nMDSGEzbHyvCeJTwhhBCvPfJL4hBAGaQwLkUriE0IYzLrTniQ+IYQxrDzzSeITQhhIFiIVQtggK7/EJ4lPCGE4K897kviEEIapXojUmkniE0IYzMrzniQ+IYThrDzvSeITQhjIgIVILZUkPiGEweRxFiGEzbH2a3yyArMQwiAKYKfot9Xm8uXLjB8/npEjRxIUFMSaNWsAKCwsJDIykoCAACIjIykqKgJAVVUWLVqERqNh1KhRnDx50ujvIIlPCGEERc/tzuzt7ZkzZw47d+7k888/59NPP+WXX34hLi4OX19fkpKS8PX1JS4uDoDU1FTOnj1LUlISCxcuZP78+UZHL4lPCGEwU9Tc8PT0xMfHB4CWLVvSqVMncnJySE5OJiwsDICwsDD27NkDoNuvKAq9e/emuLiY3Nxco+KXa3xCCIOZqq5utQsXLvDjjz/Sq1cv8vLy8PT0BKB169bk5eUBVXV1vb29dZ+prrlbfawhJPEJIQym782NuurqAly7do3o6GhefvllWrZs+bt2FLPMEpGhrhDCIFXDWEWvrS7l5eVER0czatQoAgICAPDw8NANYXNzc3F3dweq6upmZ2frPvvbmruGksQnhDDY3d/aqLpL+8orr9CpUyciIyN1+/39/UlISAAgISGBYcOG1divqiqZmZk4OzsbNcwFKS8p7pKUl7QupigvWa6tpOC6fn+ons5N7vheRkYGTzzxBF27dsXOrqoPNmPGDB588EGmTZvG5cuXadu2LbGxsbRq1QpVVVmwYAH79+/H0dGRxYsX07NnT6O+gyQ+cVck8VkX0yQ+lUI9E19rZ8u8jWCZUQkhLJuVz9yQxCeEMJiV5z1JfEIIw1RNWbPu1CeJTwhhMCvPe/I4ixDC9kiPTwhhGD3m4Vo6SXxCCIPJQqRCCJtSVWWtoaO4O5L4hBAGk8QnhLA5MtQVQtgc6fEJIWyOlec9SXxCCCNYeeaTxCeEMIiiKHVWULN0FrUslRBC1AeZsiaEsDmS+IQQNkcSnxDC5kjiE0LYHEl8QgibI4lPCGFzJPEJIWyOJD4zSk1NJTAwEI1GQ1xcXEOHI/Tw0ksv4evrS3BwcEOHIsxIEp+ZaLVaFixYwKpVq9ixYwfbt2/nl19+aeiwRB3Cw8NZtWpVQ4chzEwSn5mcOHGCe++9lw4dOtC0aVOCgoJITk5u6LBEHfr374+rq2tDhyHMTBKfmeTk5ODt7a177eXlRU5OTgNGJISoJolPCGFzJPGZiZeXF9nZ2brXOTk5eHl5NWBEQohqkvjMpGfPnpw9e5asrCzKysrYsWMH/v7+DR2WEAJZlsqsUlJSWLx4MVqtljFjxvDMM880dEiiDjNmzODo0aMUFBTg4eHB1KlTiYiIaOiwhIlJ4hNC2BwZ6gohbI4kPiGEzZHEJ4SwOZL4hBA2RxKfEMLmSOKzIt27dyc0NJTg4GCio6O5ceOG0eeaM2cOX331FQCvvPJKrQsopKWlcezYMYPb8Pf3Jz8/X+/9v9WnTx+D2lqxYgUff/yxQZ8RtksSnxVp3rw5iYmJbN++nSZNmrBhw4Ya71dUVBh13piYGLp06XLH948ePcrx48eNOrcQlkgKilupfv368dNPP5GWlsY777yDi4sLZ86cYefOnSxbtoyjR49SVlbGE088weOPP46qqixcuJCDBw/Spk0bmjRpojvX+PHjefHFF+nZsyepqaksX74crVaLm5sbMTExbNiwATs7O7Zu3crcuXPp1KkT8+bN49KlSwC8/PLL9O3bl4KCAl544QVycnLo3bs3+jwi+uyzz5KdnU1paSlPPfUU48aN0723ePFiDh48yD333MPy5ctxd3fn/PnzvPbaaxQUFNC8eXMWLlxI586dTf8LFo2bKqxG7969VVVV1fLycnXKlCnq+vXr1SNHjqi9evVSz58/r6qqqm7YsEF9//33VVVV1dLSUnX06NHq+fPn1V27dqkTJkxQKyoq1OzsbLVv377ql19+qaqqqj755JPqiRMn1Ly8PHXw4MG6cxUUFKiqqqrvvvuuumrVKl0cM2bMUNPT01VVVdWLFy+qw4cPV1VVVRcuXKiuWLFCVVVV/frrr9WuXbuqeXl5t3yPoUOH6vZXt3Hjxg01KChIzc/PV1VVVbt27aomJiaqqqqqK1asUF977TVVVVX1qaeeUs+cOaOqqqpmZmaq48ePv22MQtRGenxW5ObNm4SGhgJVPb6xY8dy/PhxevbsSYcOHQA4ePAgP/30E7t27QKgpKSEc+fOkZ6eTlBQEPb29nh5eTFw4MBbzp+ZmUm/fv1052rVqtVt4zh06FCNa4JXr17l2rVrpKen89577wHw6KOP6rWu3bp169i9ezcAly9f5ty5c7i5uWFnZ8fIkSMBCA0N5bnnnuPatWscP36c559/Xvf5srKyOtsQ4vck8VmR6mt8v+fk5KT7WVVV/vGPfzBo0KAax6SkpJgsjsrKSv7973/TrFmzuzpPWloahw4d4vPPP8fR0ZHx48dTWlp622MVRUFVVVxcXG77OxDCEHJzo5Hx8/Pjs88+o7y8HIAzZ85w/fp1+vfvz5dffolWqyU3N5e0tLRbPtu7d28yMjLIysoCoLCwEIAWLVpw7dq1Gm2sW7dO9/rHH38EqlYv3rZtG1CVaIuKimqNtaSkBFdXVxwdHTl9+jSZmZm69yorK3W91m3bttG3b19atmxJ+/bt+fLLL4GqJH/q1CnDfkFCIImv0YmIiKBLly6Eh4cTHBzMq6++ilarRaPRcO+99zJy5Ehmz55N7969b/msu7s7CxYsYOrUqYSEhDB9+nQAhg4dyu7duwkNDSUjI4NXXnmF77//nlGjRjFy5Eg+++wzAKKiosjIyCAoKIjdu3fTtm3bWmMdPHgwFRUVjBgxgrfeeqtGTE5OTpw4cYLg4GCOHDlCVFQUAG+++SabNm0iJCSEoKAg9uzZY6pfnbAhsjqLEMLmSI9PCGFzJPEJIWyOJD4hhM2RxCeEsDmS+IQQNkcSnxDC5kjiE0LYnP8PGIAif84+GmYAAAAASUVORK5CYII=\n",
            "text/plain": [
              "<Figure size 432x288 with 2 Axes>"
            ]
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "5n7uae2CQXih"
      },
      "source": [
        "# Conclusão\n",
        "\n",
        "O churn é algo que todas as empresas sofrem, até mesmo as grandes empresas. Porém, algo que pode ajudar a diminuir o churn e consequentemente alavancar o negócio é o churn prediction. \n",
        "\n",
        "Portanto, para lidar e amenizar o churn criei alguns modelos de Machine Learning a fim de prever o churn, de forma que a empresa possa contornar a situação e diminuir a sua taxa de clientes que param de usar o serviço.\n",
        "\n",
        "O modelo que obteve os melhores resultados foi definitivamente o **modelo de árvore de decisão XGBoost** que foi treinado com os dados que foram **balanceados com o método SMOTE**. \n",
        "\n",
        "Esse modelo obteve:\n",
        "\n",
        "* **405 previsões de churn** - o modelo que mais fez previsões corretas.\n",
        "\n",
        "* **62 previsões de falsos negativos** - o modelo que menos deixou passar possíveis churns.\n",
        "\n",
        "* **583 previsões de falsos positivos** -  foram feitas diversas previsões de falsos positivos. Porém, a empresa não teria problema em checar essas previsões, gerando aquelas ligações indesejadas que todos conhecemos."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "yibvFdgNQbqk"
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    }
  ]
}