Volume 6, Issue 3, March – 2021 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Artificial Neural Network Prediction Model for

Anxiety and Depression of an Individual on the
Impact of COVID-19 Lockdown in Ethiopia
Sebahadin Nasir 1*, Md Nasre Alam2, Anteneh Tiruneh 3, Demeke Getaneh 4
Woldia University, Department of Computer Science, Woldia, Ethiopia

ABSTRACT:- COVID-19 Lockdown causes different students and people affected by depression and anxiety and
health problems in the society of Ethiopia. Among these commit suicide in the pandemic [3]
problems mental health problem such as anxiety,
depression, panic and fear are common. This research Depression is a disorder that affects how you feel,
aimed to redesign a neural network model of an anxiety behave and the way how you think. Depression is a bad
and depression based on Hospital Anxiety and mood that causes sadness and loss of interest and it can lead
Depression Score (HADS) measurement techniques. We to different physical and emotional problems. Someone who
collected 713 data from different individuals including affected by depression may not able to perform his day to
students, working and non-working male and female age day activities, and occasionally he/she may fell as if life
group from 16 to 55 using online survey. In order to isn’t worth living [4]. Nowadays depression is foremost
online survey, we prepared 7 questions using HADS contributor to the worldwide burden of disease and it affects
standard for anxiety and depression. Each of these people in any societies through the world. In 2012,
questions has four answer scores from 0 to 3. We depression is estimated to affect 350 million people [5].
generate neural network model on the basis of Environmental factors, personality, genetics and
participant response and HADS measurement technique biochemistry are some of the factors of depression [6].
in order to classify the level of Anxiety and Depression. According to oxford dictionary anxiety is a strong desire or
The level of anxiety and depression can be normal, mild, concern to do something or for something to happen and it is
moderate and severe. The model was tasted and its also defined as a feeling of worry, nervousness, or unease
specificity was 0.997940975 for anxiety and 0.996577687 about something with an uncertain outcome [7]. There is
for depression. We achieved the sensitivity value for different classification of anxiety and there are different risk
anxiety is 0.926666667 and for depression is factors of anxiety among this stress due to an illness, stress
0.945205479. We compared the model accuracy buildup, Trauma, other mental health disorder and drugs or
manually using HADS technique. We found the Average alcohol [8]. Feeling faint or dizzy, dry mouth, sweating,
Percentage Value (APV) 0.017379846 and 0.018365 for apprehension and worry, restlessness, distress and fear are
anxiety and depression respectively. This study can some of the symptoms of an anxiety attack [9]. The causes
further designed to recommend some advices on what an of both anxiety and depression are multifactorial, including
individual may do or what kind of measurements they biological, economic, social, environmental and cultural.
must do in each level of Anxiety and depression. Diagnosis of it is made by psychiatrists or psychologists
according to Diagnostic and Statistical Manual of Mental
Keywords:- Anxiety, Depression, COVID-19, Artificial Disorders (DSM)-5 [10] or International Classification of
Neural Network, Hospital Anxiety and Depression Scale. Diseases (ICD) 10 [11].

I. INTRODUCTION Machine-learning approaches is better than traditional

statistical prediction models on recognizing multipart
COVID-19 is an illness which stands for corona virus contours and patterns in data, and preventing hypothetically
disease that caused by a virus that can spread from person to imprecise model specifications [12]. Machine learning and
person. As World Health Organization (WHO) reported that data mining methods is used in health centers to diagnose
the COVID-19 was first identified in China, Wuhan city and complex diagnosis based on the patient’s history and this
spread throughout the world. In 2021 more than 80,000,000 can be used for making logical decisions. Diagnosis of
cases reported [1]. According to the new survey done by anxiety disorders is a very complex and challenging task.
WHO, the COVID-19 pandemic has disordered mental Thus using machine learning, an anxiety can be diagnosis
health service demand is increasing in the society which is with the high accuracy [13]. Articial Neural Networks
93% of countries worldwide. From depression and anxiety (ANNs) are densely interconnected and adaptive processing
alone approximately 1 trillion united states dollar economic units, with an inherent ability for learning from experience
productivity is lost in each year worldwide [2]. In Ethiopia, and discovering new knowledge [14]. Hospital anxiety and
more than 26 million students were distorted during the depression score (HADS) is a measurement tool for clinical
COVID -19 pandemic and it is reported in news a number of practice and research. It was designed to measure anxiety
and depression in general patients in health centers [15]. It is

IJISRT21MAR692 www.ijisrt.com 1221

Volume 6, Issue 3, March – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
simple and easy to use calculate the level of anxiety and learning algorithms. They used python programming with
depression of an individual’s[16]. CNN for modeling experiments that scores accuracy and
recall of 78% and 0.72, respectively.
Different authors conduct a research on machine
learning model for prediction of anxiety and depression. II. METHOD AND EXPERIMENT
Lana G. Tennenhousea and et al. [17] did a study on
machine-learning models for depression and anxiety in 2.1 Data Collection
individuals with immune-mediated inflammatory disease. For this study, the data was collected from 1607
They used Patient-reorted out-come measures (PROMs) to individuals using Google forms. Among these 1607
predict anxiety and depression of an individual with individuals who filled the online form 778 are females and
immune-mediated inflammatory disease (IMID) using 829 were male, age between 16- 55, it includes students,
machine learning. They collected 637 data to train the employed, unemployed and household. In the online form
machine, logistic regression (LR), neural network (NN), and 14 questions where used in the questioner, 7 for anxiety and
random forest (RF) technique was used. However, their 7 for depression. The questioner used was based on HADS,
model was designed anxiety and depression caused by Hospital anxiety and depression scale and edited according
deregulation of the immune system. to Ethiopian context [18]. The question used for Anxiety
and depression used in the online questioner is listed below:
Anu priyaa and et al. designed a model that predicting
anxiety, depression and stress in modern life using machine

Anxiety Depression
I feel tense or 'wound up'? I still enjoy the things I used to enjoy?
I get a sort of frightened feeling as if something awful is about I can laugh and see the funny side of things?
to happen? I feel cheerful?
Worrying thoughts go through my mind? I feel as if I am slowed down?
I can sit at ease and feel relaxed? I have lost interest in my appearance?
I get a sort of frightened feeling like 'butterflies' in the I look forward with enjoyment to things?
stomach?
I feel restless as I have to be on the move? I can enjoy a good book or radio or TV program?
I get sudden feelings of panic?

The expected answer and score point for each question in the questioner for Anxiety and depression is shown below in Table
1[18].

Table 1. Anxiety and depression questioner answer and score.

Anxiety Depression
Answer Score Answer Score
Very definitely and quite badly 3 Not at all 0
Yes, but not too badly 2 Sometimes 1
A little, but it doesn't worry me 1 Very often 2
Not at all 0 Nearly all the time 3

The collected data for Anxiety and depression score level of individual was categorized as normal, mild, severe and
moderate using HADS technique [18]. The four categories such as normal, mild, severe and moderate are coded using decimal
numbers as 1, 2, 3 and 4 respectively and these used for training in the neural network. In Figure 1 the questioner collected from
each person is coded from p1-Ans to p1607-Ans and the questions are coded horizontally with their respective answer score
vertically.

IJISRT21MAR692 www.ijisrt.com 1222

Volume 6, Issue 3, March – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

(a) Anxiety (b) Depression

Figure 1: Anxiety and Depression collected questioner answer of an individual

2.2 Experiment
MatLab programming language was used to train and 2.4 Training Neural Network
test the model generated. It is a High-level language for The multilayer feed forward network can be trained for
technical computing and interactive environment used to function approximation pattern recognition [20]. In
perform computationally intensive tasks faster that other implementing the training there are two modes such as
traditional programming languages like c and c++ [19]. The incremental mode and batch mode. In incremental mode, the
network is created to recognize the Pattern using gradient is calculated and the weights are updated after
feedforward networks that can be trained to classify inputs every single input is applied to the network. In batch mode,
(each score/answer of the questions shown in Figure 1) before the weights are updated all the inputs in the training
according to target classes (anxiety and depression class set are applied to the network. In using the Neural Network
calculated using HADS). With the help of pattern network Toolbox software, batch training is significantly faster and
20 hidden layers are used as shown in neural network design produces smaller errors than incremental training in most
Figure 2 below. problems [21] .

2.3 Dividing Data The Scaled Conjugate Gradient algorithm is used by

In multilayer networks training, the 712 data is divided default by pattern recognition. The network is trained using
into three subsets as training, validation and testing sets. In Scaled conjugate gradient back propagation algorithm using
the first subset, among all data 70% is used for training, the input, target and the created network parameters. There
which is used for computing the gradient and updating the are 7 inputs and 1 output for both anxiety and depression
network weights and biases. The second subset is the training as shown in Figure 2 below. The training was
validation set and here we used 15% of all data. The third stopped at epoch 54, it takes 0.22 seconds long, with the
subset is testing and similarly with the validation 15% of all performance 0.664, at 1.29 performance gradient and with 6
data is used. validation performance.

IJISRT21MAR692 www.ijisrt.com 1223

Volume 6, Issue 3, March – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Figure 2: The design of Artificial Neural Network

2.5 Training performance validation and testing

The training performance shows the final mean square error is small at epoch 54 as shown in Figure 3. In the figure the
validation and test set error have similar characteristics and it achieves the best validation performance 0.049968 at iteration 48
and no significant over fitting occurred at iteration 48.

Figure 3. Training performance validation and testing

III. RESULTS AND DISCUSSIONS

The accuracy result of the model is compared with the manually calculated value by hospital anxiety and depression score.
The anxiety and depression error rate measured in absolute percentage value (APV) of 0.017379846 and 0.018365 respectively.
Moreover the sensitivity is calculated for anxiety and depression using APV as 0.926666667 and 0.945205479 respectively, the
specificity of the anxiety and depression is 0.997940975 and 0.996577687 respectively.

IJISRT21MAR692 www.ijisrt.com 1224

Volume 6, Issue 3, March – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Table 2: Result of the designed model of an anxiety and depression
True Class(Anxiety and depression ) using HADS
Class Precision Sensitivity Specificity F1-score Accuracy Error Rate
Anxiety Normal 0.995271868 0.976798144 0.994630872 0.985948 0.985065 0.014934661
Mild 0.916363636 0.984375 0.95481336 0.949153 0.964706 0.035294118
Moderate 0.968023256 0.982300885 0.991324921 0.97511 0.989421 0.010578718
Severe 0.978873239 0.926666667 0.997940975 0.952055 0.991288 0.008711886
Predicted class using ANN

APV 0.964633 0.967535174 0.984677532 0.9655665 0.98262 0.017379846

Depression Normal 0.985176739 0.998843931 0.982479784 0.991963 0.991288 0.008711886
Mild 0.964705882 0.917910448 0.981052632 0.940727 0.958277 0.041722746
Moderate 0.957831325 0.969512195 0.989053948 0.963636 0.985065 0.014934661
Severe 0.965034965 0.945205479 0.996577687 0.955017 0.99191 0.008089608
APV 0.968187 0.957868 0.987291 0.962836 0.981635 0.018365

IV. CONCLUSION conditions/depression/symptoms-causes/syc-

20356007, last accessed on May 2020.
In this study, our main finding shows neural network [5]. DEPRESSION, A Global Public Health Concern,
model for prediction of anxiety and depression caused by Developed by Marina Marcus, M. Taghi Yasamy,
COVID -19 lockdown in Ethiopia was designed and gives Mark van Ommeren, and Dan Chisholm, Shekhar
us a good result. During the model generation we collected Saxena WHO Department of Mental Health and
713 data using online survey in Ethiopia and in answering Substance Abuse, 2012
the questioner employed, unemployed, student, female male [6]. American Psychiatric Association, Risk Factors for
and age between 16 up to 55 was participated. Based on the Depression, available at
standard HADS, 7 question used form both anxiety and https://www.psychiatry.org/patients-
depression. In the experiment, feedforward neural network families/depression/what-is-depression last accessed
was used and it achieves a best accuracy of 0.049968 at on July 2020.
iteration 54. The overall result of specificity and sensitivity [7]. Definition from oxford dictionary,
in detection of Anxiety was 0.997940975 and 0.926666667 https://languages.oup.com/google-dictionary-en/
respectively and the overall specificity and sensitivity result [8]. Anxiety disorders, Available at
of depression was 0.996577687 and 0.945205479 https://www.mayoclinic.org/diseases-
respectively. The result of the neural network model was conditions/anxiety/symptoms-causes/syc-20350961,
compared with the ground truth value calculated manually Last accessed on March 02, 2020
using HADS and the average percentage error of the model [9]. Everything You Need to Know About Anxiety,
in anxiety and depression detection is 0.017379846 and https://www.healthline.com/health/ anxiety#anxiety-
0.018365 respectively. In the future we will further consider attack
attributes like age, income level and gender in generating [10]. American Psychiatric Association Force DSMT.
the model. Diagnostic and statistical manual of mental disorders:
DSM-5. American Psychiatric Association. 2016;
REFERENCES 2013.
[11]. World Health Organization. The ICD-10 classification
[1]. Worldometers, “corona virus updates”, available at of mental and behavioural dis-orders: diagnostic
https://www.worldometers.info/ last accessed on criteria for research. 1993.] [A Neural Network
January, 2021 Based M o del for Predicting Psychological Conditions
[2]. World Health Organization, “ COVID-19 disrupting [12]. Lana G. Tennenhousea, Ruth Ann Marrie, Charles N.
mental health services in most countries, WHO Bernstein, Lisa M. Lix "Machine-learning models for
survey”, available at https://www.who.int/news/, Last depression and anxiety in individuals with immune-
accessed on June, 2020. mediated inflammatory disease" Journal of
[3]. UNICEF, “The case for safely reopening schools in Psychosomatic Research 134, 110126, 2020.
Ethiopia”, Available at [13]. Emmanuel G. Pintelas, Theodore Kotsilieris, Ioannis
https://www.unicef.org/ethiopia/stories/case-safely- E. Livieris and Panagiotis Pintelas, “A review of
reopening-schools-ethiopia, Last accessed on machine learning prediction methods for anxiety
November, 2020. disorders” , ResearchGate, DOI:
[4]. Depression (major depressive disorder), Mayo Clinic 10.1145/3218585.3218587, Conference Paper · July
available at https://www.mayoclinic.org/diseases- 2018

IJISRT21MAR692 www.ijisrt.com 1225

Volume 6, Issue 3, March – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
[14]. C.M. Bishop. 1995. Neural Networks for Pattern [18]. Hospital Anxiety and Depression Scale (HADS),
Recognition. Oxford. Availabel at
[15]. Zigmond AS, Snaith RP. The Hospital Anxiety and https://www.svri.org/sites/default/files/attachments/20
Depression Scale. Acta Psychiatr Scand 1983; 67:361– 16-01-13/HADS.pdf Last Accessed on March 06,
370. 2020.
[16]. McManus S, Meltzer H, Brugha T, Bebbington P, [19]. University of Birmingham, “About
Jenkins R, eds. Adult Psychiatric Morbidity in MATLAB”Available at, https://intranet.birmingham
England 2007. Results of Household Survey. .ac.uk/collaboration/hpc-
http://www.hscic.gov.uk/catalogue/ PUB02931/adul- research/matlab/about/index.aspx#:~:text= MATLAB.
psyc-morb-res-hou-sur-eng-2007-rep.pdf (February , Last Accessed on March 1, 2021.
2014, date last accessed). [20]. Feedforward Neural Network, B.K. Lavine, T.R.
[17]. Lana G. Tennenhousea and et al. [ ] Lana G. Blank, in Comprehensive Chemometrics, Volume 3,
Tennenhousea, Ruth Ann Marrie, Charles N. ScienceDirect, 2009
Bernstein, Lisa M. Lix "Machine-learning models for [21]. MathWorks, "Train and Apply Multilayer Shallow
depression and anxiety in individuals with immune- Neural Networks" Available at
mediated inflammatory disease" Journal of https://www.mathworks.com/help/deeplearning/
Psychosomatic Research 134, 110126, 2020. ug/train-and-apply-multilayer-neural-networks.html,
Last Accessed on March 1, 2021.

AUTHORS

Sebahadin Nasir Shafi is graduated bsc in computer science from Ambo University in 2012 and he
received his msc in computer science from Addis Ababa university department of science in 2017.
Currently he is working as lecturer and researcher at Woldia university department of computer
science. Moreover he is working on position of an associate registrar at Woldia university technology
Institute. He is working and interested in research areas of computer vision, digital image processing
and patterns recognition, machine learning and Artificial intelligence wireless networking.

Md. NasreAlam received his bachelor in computer application (bca) from m.c.r.p university, bhopal,
india in 2004, and m.sc. In computer science from hamdard university, new delhi, india in 2007. He
did phd at the graduate school of it and telecommunication engineering in inha university, incheon,
south korea. From 2014 to 2017 he worked as post doctor in chonbuk national university, jeonju,
south korea. Now he is serving as assistant professor at department of computer science, Woldia
university, Ethiopia. His research interests are wireless sensor networks, wireless communications,
wireless ad- hoc networks, wireless body area networks and wireless personal area networks.

Anteneh Tiruneh Terefe is MSc graduate from Adama Science and Techology University in School of
Computing in 2017. Currently he is working as lecturer and researcher at Woldian University Institute
of Technology Department of Computer Science and he is interested in Internet of things and data
science research.

Demeke Getaneh Mergia received his bachelor degree in Software Engineering from Adama
University, Ethiopia in 2013 and MSc. Degree in Software Engineering from Adama University in
2018. Since 2014 up to 2016 he is working as Assistant lecturer at institute of technology, School of
Computing, department of computer Science in Woldia University, Ethiopia. Since March 2018 he
has been working as a Lecturer in Woldia University. Besides teaching he has been doing researches,
technology transfer projects, and community services. His area of interest includes software
engineering, wireless sensor networks, wireless Ad Hoc network, AI, Machine Learning and Data
Science.

IJISRT21MAR692 www.ijisrt.com 1226