Mental Illness Prediction using Machine Learning Algorithms

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 10 Issue: 08 | Aug 2023 www.irjet.net p-ISSN: 2395-0072
© 2023, IRJET | Impact Factor value: 8.226 | ISO 9001:2008 Certified Journal | Page 191
Mental Illness Prediction using Machine Learning Algorithms
Falguni Wani1, Ved Deore2, Shivam Gorane3, Santosh Chobe4
1,2,3Student, Dept. of Computer Engineering, Pimpri Chinchwad College of Engineering and Research, Ravet, Pune,
Maharashtra, India
4Professor, Dept. of Computer Engineering, Pimpri Chinchwad College of Engineering and Research, Ravet, Pune,
Maharashtra, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Depression is one of the most concerned issues in
the society and it is not limited to certain age of a person.
Depression management is an approach for analyzing and
working on these concerns and lead to quality of life. The idea
behind this work is to analyze depression, anxiety and stress
based on some psychological test like Depression Anxiety
Stress Scale-21(DASS 21). Machine learning is an emerging
field in computer science and has ability to predict outcome
based on certain situations or inputs. Machine learning
algorithms are used to predict depression, anxiety and stress
levels by using standard psychological scale. Training and
testing datasets are used to train and test the developed
machine learningmodel. Variousmachinelearningalgorithms
like Support Vector Machine, Random Forest, NaïveBayes, etc.
are implemented and compared in order to evaluate the best
among all. The accuracy of the best algorithm isboostedusing
the boosting technique of ensemble learning method and a
user interface is used for self-evaluation. From the
classification algorithms used SVM has surpassed the other
machine learning algorithms and then it is boosted using
AdaBoost giving highest accuracy for prediction.
Key Words: Depression, Anxiety, Stress, Classification,
Depression Anxiety Stress Scale-21(DASS-21), Supervised
Learning, AdaBoost, Support Vector Machine
1. INTRODUCTION
Healthcare is among the serious issues in front of
the entire world regardless of any circumstances.Asa ruling
interest globally, besuited, well organized, effective and
robust wellness systems are built to improve and conserve
the quality standards of life. Anxiety, depression, stress,
irritation and disappointment have become so normal that
individuals now imagine them to be part of personal and
professional life.
The World Health Organization (WHO) has
estimated that 3.8% of the population experience
depression, including 5% of adults (4% among men and 6%
among women), and 5.7% of adults older than 60 years.
Approximately 280 million people in the world have
depression [1]. Differentiating between anxiety and
depression is complicated formachines;therefore,a suitable
machine learning algorithm is necessary for an applicable
recognition.
Mental health is an integral and essential
component of health. The WHO constitution states: "Health
is a state of complete physical, mental and social well-being
and not merely the absence of disease or infirmity."[2]. The
leading symptoms of depression from a medical point of
view are lack of concentration, loss of memory, loss of
interest in recreational activities, an inability to make
decisions, overeating and weight gain, weight loss, low
appetite and irritation, etc. These symptoms have a
significant effect on crucial areas of an individual’s life.
The symptoms of anxiety are irritability, insomnia,
nervousness, sweating, fatigue, panic, increased heart rate
and a sense that something is about to happen, difficulty in
concentrating and rapid breathing.
The common symptoms of stress are low energy
levels, feeling upset or agitated, chronic headaches,
impotence to relax, recurring overreaction and persistent
colds or infections. Thus, anxiety,stressanddepressionhave
many common symptoms including fatigue, chest pain,
insomnia, inability to concentrate and increased heart rate
all of which makes classification tough for machines.
This paper is structured as follows: Section 2
explores related studies on anxiety, depression and stress
along with the methods and techniques that were adopted.
Section 3 describes the dataset used in the research herein,
while Section 4 discusses the various classification
algorithms. Section 5 studies the researchgapfound.Section
6 includes experimental setup used to perform this study,
while section 7 describes the proposed system. Section 8
compares the resultsofmachinelearningalgorithms.Finally,
section 9 is the conclusion, which summarizes the study in
its entirety.
2. LITERATURE REVIEW
The literature survey shows the study of various
machine learning algorithms to predict depression, anxiety
and stress.
In [3], Anu Priya, et. al. proposed machine learning
model for predicting different levels of depression, anxiety
and stress. They applied different machine learning
algorithms like Support Vector Machine (SVM), Decision
Tree (DT), Naïve Bayes (NB), Random Forest (RF) and K-
Nearest Neighbors (KNN). They also calculated different

comparison factors for choosing the best algorithm and
found out that Naïve Bayes algorithm gives the best results.
They used the Depression, Anxiety and Stress Scale
questionnaire (DASS-21) to train and test their model. The
size of dataset was less and there wereimbalancedclassesin
the confusion matrix, so the model did not give the expected
accuracy and the decision was made on the f1-scorescore so
Random Forest was chosen to be the best fit for all three
classes.
Fig -1: Accuracy chart for [3]
Astha Singh et. al. [4] proposed a model for
identification of anxiety and depression. They collected the
data by using standard DASS-21 questionnaire. They used
some of the standard ML algorithms like decision tree (DT),
SVM, Naïve Bayes, Random Forest along with KNN for
training and testing purpose. Although the accuracy wasnot
more than 95%, they selected SVM classifier as the best
among all other. They also faced some problems related to
dataset and affected its accuracy.
Fig -2: Accuracy Chart for [4]
Hritik Nandanwar et. al. [5] designed a model for
depression prediction. The dataset used by them was
collection of tweets from Twitter. They compared the
performance of different machine learning models with
labelled Twitter dataset. Different evaluationmetricslikef1-
score, recall and precision have been used to compare the
performance. They got better results using Bag of Words
with AdaBoost classifier.
Ruihu Wang [6] surveyed AdaBoost classifier for
classification, feature selection, and their relation with
support vector machine. They studied the fundamentals
about AdaBoost algorithm for feature extraction and
selection. They found out that AdaBoost algorithm gives
better results and it has been widely used in many real time
applications. The study also showed that one of the
optimization algorithms known as Particle Swarm
Optimization (PSO) also gives good results for prediction.
S Samanvitha, et. al. [7] built the model for
depression detection using text data. They took data from
different social media for building the model. As it is seen
that people often express their feelings on online platforms.
They tested their model with algorithms like Logistic
Regression, Naïve Bayes, Random Forest and SVM classifier.
They concluded that Naïve Bayes classifier gives the best
results.
Paphaychit Bounkeomany [8]designeda systemfor
depressiondetectionusingspeech.TheyusedAdaboost-ELM
framework which uses random numbersastheyarenumber
of meta-samples of dictionary atoms. They also enhanced
this model using random dynamic integrated weighted
classification model.
Ananna Saha et. al. [9] proposed a machinelearning
model of sentiment analysis of depressed person. They took
the dataset of user generated contents from different social
media applications like Facebook, Twitter. They have used
python textblob package for different sentimentlevels.They
implemented Random Forest, Naïve Bayes classifier, DT,
Sequential Minimal Optimization, Support Vector Machine
(SVM), boosting technique such as Adaboost, Logistic
Regression, Bagging, Multilayer Perceptron and Stacking
algorithms. Among all they got 60.54% with Random Forest
Algorithm.

Heidi Mochari-Greenberger et. al. [10] compared
different psychological scales. That includeDASS-21,GAD-7,
PHQ-8. Their result of study shows that how GAD-7 and
DASS-21 scale categorize severity of symptoms withrespect
to each other in population with behavioral and medical
conditions of health. Inthisstudy,theauthorsconcludedthat
DASS-21 remains the best scale to use as it consists of least
number of questions to be answered.
Anju Prabha et. al. [11] used the Stroop Test
mechanism to identify depression among people in COVID-
19 situation. They found that there was a visible difference
between the mental conditions of the patients between a
certain time stimuli. They used machine learningalgorithms
such as Support Vector Machine (SVM), XGBoost, etc. to
calculate the accuracy of the data and found that XGBoost
algorithm gave the highest accuracy for the dataset. The
XGBoost algorithm gave accuracy of 85.71% as compared to
other algorithms used.
Shivangi Yadav et. al. [12] used Machine Learning
and employed a wide range of Machine Learning algorithms
to predict depression in people. They collected data by
questioning people about their home, workplace
environment and family history, etc. They used Machine
Learning algorithms such as: K-Nearest Neighbors (KNN),
Decision Tree, Multinomial Logistic Regression, Random
Forest Classifier, Bagging, Boosting and Stacking. The best
performance statistics was shown by boosting algorithm
which gave accuracy of 81.75% which was then followed by
Random Forest Classifier with accuracy of 81.22%.
Md. Mehedi Hassan et. al. [13] have developed
prediction models by classifying the dataset related to
depression which were taken from Kaggle. They had
primarily focused on feature selection. They selected the
features after preprocessing the data and applied Logistic
Regression, Correlation Matrix and Decision Tree methods.
They applied different Machine Learning algorithms suchas
Logistic Regression, K-NN,SVM,andNaïveBayesforbuilding
and classifying models. They got the best classification and
accuracy for K-NN which is 79%. The other algorithms such
as Logistic Regression, SVM, and Naive Bayes showed an
accuracy of 77%, but K-NN was selected to be the best fit.
G H Suhas, et. al. [14] identified the risk of
depression among people in the form of text. They collected
and analyzed sentences from people to predict or detect
whether the person is suffering through depression or not.
They used different Machine Learning algorithms andfound
that Random Forest classifier gives the best accuracy when
compared to CNN. Their system takes input frompeopleand
predict based on the responses.
Aanchal Bisht et. al. [15] proposed a methodology
that will help teachers and parents to predict the levels of
stress which students experience. They surveyed school
children with a variety of 26 questionstoanalyzetheirstress
levels and cure those using Machine Learning algorithms.
They used different Machine Learning algorithms such as
Decision Tree, Logistic Regression, K Nearest Neighborsand
Random Forest. They found that K NearestNeighbors(KNN)
gave accuracy of 88% and proved to be the best for the
implementation.
Akshada Kene et. al. [16] presented a study on
previous research on stress detection using Machine
Learning algorithms. They used the PhysioBank dataset to
analyze different stress levels. They used statistical analysis
for feature selection and extraction and found that gradient
boost algorithm to be successful on the dataset used. The
results demonstrated that the model displayed the accuracy
of 83.33%, specificity of 75%, Sensitivity of 75%, Positive
Recall value of 90%, and many more. The machine learning
algorithms such as KNN, Random Forest, SVM and Naïve
Bayes were used. Authors claimed Naïve Bayes model to be
effective and efficient for stress classification andprediction
with accuracy of 88%.

Somnath Sinha et. al. [17] have implementedStress
Prediction for the students and staff in the university
premises to check whether they are stressed or stress-free.
They used Machine Learning algorithms such as K Nearest
Neighbors and Naïve Bayes to predict the results. They
compared both the algorithms and tested them with easy
manner. They concluded that Naïve Bayes algorithmismore
efficient than KNN and has a high efficacy rate. The authors
claimed the accuracy level over 94 percent for Naïve Bayes
whereas the accuracy level for KNN as 87 percentage.
Anika Kapoor et. al. [18] in their research aimed to
identify the anxiety disorders using Machine Learning
Techniques. They identified symptoms such as Generalized
Anxiety Disorder (GAD), Panic Disorder (PD), Post-
Traumatic Stress Disorder, etc. They collected the dataset
from many organizations/institutions/hospitals,etc.mainly
through surveys and questionnaire related to the disease.
For prediction they used Machine Learning algorithms such
as Random Forest, Linear Regression, Support Vector
Machine and others. Finally,theyconcludedthatSVMhas the
highest accuracyforGeneralisedAnxietyDisorder(GAD)and
Social Anxiety Disorder (SAD) while it was left behind by a
margin of just 0.4% by GB Decision Tree (DT) for Post-
Traumatic Stress Disorder (PTSD) and by 1% for Obsessive-
Compulsive Disorder (OCD) by Random Forest (RF) which
also achieved the best accuracy for Panic Disorder (PD).
Ahnaf Atef Choudhury et. al. [19] in their approach
proposed predicting depression in university
undergraduates and recommend them to the psychiatrist.
They collected data from the after consultation with
counselors, professors and psychologists.Theauthorsfound
Random Forest to be the best algorithm followed by the
Support Vector Machine (SVM) with accuracy around 73%
and KNN with accuarcy 60% respectively. The Random
Forest algorithm gave a better precision, recall andlowfalse
negatives. This research aimedtopredictdepressioninearly
stages and ensure quick recoveryforthevictimstoavoid any
further mishaps.
3. DATASET
The dataset consists of 21 questions based on the
DASS-21[20] questionnaire, under the categories like,
Depression, Anxiety, and Stress Scale. These questions are
divided into the set of 7 for each category and the answerfor
each question is represented as numeric text as follows:
0 – Does not apply to me.
1 – Apply to me to some degree or sometimes.
2 – Apply to me to a considerable degree.
3 – Apply to me most of the time.
Table 1 shows the questions asked under each category.
Table -1: DASS-21 Questionnaire
Depression Anxiety Stress
I couldn’t seem
to experience
any positive
feeling at all.
I was aware of
dryness of my
mouth
I found it hard to
wind down (calm
down)
I found it
difficult to
work up the
initiative to do
things
I experienced
breathing
difficulty (e.g.,
excessively rapid
breathing,
breathlessness in
the absence of
physical exertion)
I tended to over-
react to situations
I felt that I had
nothing to look
forward
I experienced
trembling
(shaking, e.g., in
the hands)
I felt that I was
using a lot of
nervous energy (an
excess of energy
that you have when
you are worried)
I felt down-
hearted and
blue (feeling
sad and
discouraged)
I was worried
about situations
in which I might
panic and make a
fool of myself
I found myself
getting agitated
(upset, disturbed)
I was unable to
become
enthusiastic
I felt I was close
to panic
I found it difficult
to relax

about anything
I felt I was not
worth much as
a person
I was aware of
the action of my
heart in the
absence of
physical exertion
(e.g., sense of
heart rate
increase, heart
missing a beat)
I was intolerant of
anything that kept
me from getting on
with what I was
doing
I felt that life
was
meaningless
I felt scared
without any good
reason
I felt that I was
rather
touchy(sensitive)
4. CLASSIFICATION
4.1 Decision Tree
Decision Tree is Supervised Machine Learning
algorithm which is used for classification as well as
regression problems, but mostly it is used for classification
problems. It is a tree structuredclassifier,wherethefeatures
of the dataset are represented as internal nodes, decision
rules are represented by the branches and outcome is
represented by each leaf node.
Fig -13: Decision Tree Example
4.2 Gaussian Naïve Bayes
Naïve Bayes classifier is a Supervised Machine
Learning algorithm based on Bayes theorem and is used for
solving classification problems. It includes training high
dimensional dataset and is one of the simple and effective
classification algorithms. It helps in building machine
learning models that can make quick predictions. It is a
probabilistic classifier and predicts results based on
probability. The formula of Bayes theorem is as follows:
P (X|Y) = P (Y|X). P (X) / P (Y)
4.3 Random Forest
Random Forest is a Supervised Machine Learning
algorithm which is used for Classification as well as
Regression problems. It is a process of combining several
classifiers to solve complex problems to improve the
performance of the model. It contains a number of decision
trees on various subsets of dataset and take the average to
improve its accuracy. It can also handle datasetthatcontains
continuous variables in regression and categorical variable
in case of classification.
Fig -14: Random Forest Example
4.4 Support Vector Machine
Support Vector Machine (SVM) is a Supervised
Machine Learning algorithm which is used for Classification
as well as Regression problems. It is mostly used as
Classification problems. It creates a decision boundary that
can segregate n-dimensional spaces classes so that we can
easily classify the data point in its correct category in future.
SVM chooses extreme points to create the hyper plane and
these extreme cases is termed as Support Vector Machine.
Fig -15: Support Vector Machine
4.5 XGBoost
XGBoost is an ensemble learning method that
combines multipleweak classifiersintoa strongerprediction
model. XGBoost is also known as “Extreme Gradient
Boosting”. It also supports for parallel processing and one of
its key features is its efficiency of handling missing values.
4.6 AdaBoost
AdaBoost is short form for “Adaptive Boosting”.Itis
an ensemble learning technique which is used to make
strong classifier based on the weak classifiers. It was first
developed for the purpose of binary classification. The
common estimator used with AdaBoost is decision treewith
one level, i.e., decision tree with 1 split. These are also called
as decision stumps.

Fig -16: Adaptive Boosting
5. RESEARCH GAP
While studying the topic it was found that there are
no boosting algorithms used with the DASS-21 scale. The
accuracy given by various algorithms can be boosted using
Adaboost or XGBoost algorithms. The boosted accuracy will
help classify the problems more accurately.
6. EXPERIMENTAL SETUP
The dataset used is based on DASS21 standard
questionnaire and the other supervised learning algorithms
in this project. The system specifications are as follows:
 Processor: Intel(R) Core(TM) i5-8265U CPU @
1.60GHz 1.80 GHz
 RAM: 8GB
 System type: 64-bit Operating System
7. PROPOSED SYSTEM
Figure 17 depicts the proposed system. In the
proposed system, the data is collected from the
users/patients. This data is based on answers provided by
the user/patient as per the standard questionnaire of
Depression, Stress and Anxiety. The data has the features of
the standard psychological factors. Next, the data pre-
processing is done through handling missing values,
transformation, encoding, etc. In the process of feature
extraction, the strong and independent features areselected
to achieve the target variable. Next, the model training is
performed on the Machine Learning Algorithms such as:
Random Forest,Support VectorMachine(SVM),NaïveBayes,
Decision Tree and XGBoost. It was found that SVM
outperformed other algorithms. ThenAdaBoostalgorithmis
used to boost the accuracy of the model. After applying the
algorithms, the severity levels of the depression, stress and
anxiety are calculated. Some tips on how to overcome the
depression will be provided to the user/patient or some
counselling may be provided. The performance of themodel
was measured with performance metrics viz. accuracy,
recall, precision, f1-score and it was observed that the
proposed system gives better results.
Fig -17: Proposed System Architecture
8. RESULT
Table 2 depicts theaccuracyfordifferentalgorithms
used for predicting severity levels ofdepression,anxietyand
stress. Proposed system yielded the highestaccuracyamong
all.
Table -2: Comparison Table
Algorithm
Accuracy
Naïve Bayes 75.81 56.38 72.54
Decision Tree 61.21 81.04 63.08
Random Forest 65.83 65.21 61.32
XGBoost 79.08 77.12 69.93
Support Vector
Machine
86.27 89.54 89.54
Proposed
System
94.08 92.89 93.49

0
10
20
30
40
50
60
70
80
90
100
Chart -1: Accuracy Comparison Chart
9. CONCLUSION
From the study, it has been analysed that there are
five levels of severity viz. Normal, Mild, Moderate, Severe
and Extremely Severe for depression,stressandanxiety. The
datasets used by various researchers were collected using a
standard questionnaire to measure the frequent symptoms.
The earlier research has shown an accuracy with single
algorithm to a satisfactory level. The proposed system
bestowed the accuracy of 94.08, 92.89, and 93.49 for
Depression, Anxiety and Stress respectively.
10. REFERENCES
[1] https://www.who.int/news-room/fact-sheets/
detail/depression”, 01/04/2023
[2] https://www.who.int/data/gho/data/major-themes/
health-and- well-being”, 01/04/2023
[3] Anu Priya, Shruti Garg, Neha Prerna Tigga, “Predicting
Anxiety, Depression and Stress in Modern Life using
Machine Learning Algorithms” in International
Conference on Computational Intelligence and Data
Science (ICCIDS 2019), 2020
[4] Astha Singh, Divya Kumar, “Identification of Anxiety
and Depression Using DASS-21 Questionnaire and
Machine Learning” in First International Conference on
Advances in Computing and Future Communication
Technologies (ICACFCT), July 2022.
[5] Hritik Nandanwar, Sahiti Nallamolu, “Depression
Prediction on Twitter using Machine Learning
Algorithms” in 2nd Global Conference for
Advancement in Technology (GCAT), November 2021.
[6] Ruihu Wang, “AdaBoost for Feature Selection,
Classification and Its Relation with SVM, A Review” in
International Conference on Solid State Devices and
Materials Science, 2012.
[7] S Samanvitha; A R Bindiya, Shreya Sudhanva; B S
Mahanand, “Naïve Bayes Classifier for depression
detection using text data” in 5th International
Conference on Electrical, Electronics, Communication,
Computer Technologies and Optimization Techniques
(ICEECCOT), February 2022.
[8] Paphaychit Bounkeomany, “Speech Major Depression
Detection Based on Adaboost-ELM Algorithm” in IEEE
International Conference on Information Technology,
Big Data and Artificial Intelligence (ICIBA), December
2020.
[9] Ananna Saha, Ahmed Al Marouf, Rafayet Hossain,
“Sentiment Analysis from Depression-Related User-
Generated Contents from social media” in 8th
International Conference on Computer and
Communication Engineering (ICCCE), July 2021.
[10] Heidi Mochari Greenberger, Reena L Pande, Aimee
Peters, Lila Peters, Evie Andreopoulos, Naomi Pollock,
“Comparison of DASS-21, PHQ-8, andGAD-7ina virtual
behavioral health care setting”.
[11] Anju Prabha, Jyoti Yadav, Asha Rani, Vijander Signh, “A
Pilot study for Depression Detection during COVID-19
using Stroop Test” in 8th International Conference on
Signal Processing and Integrated Networks (SPIN),
October 2021.
[12] Shivangi Yadav, Tanishk Kaim, Shobhit Gupta,
“Predicting Depression from RoutineSurveyData using
Machine Learning” in 2nd International Conferenceson
Advances in Computing, Communication Control and
Networking, March 2021.
[13] Md. Mehedi Hassan, Md. Asif Rakib Khan, Khan Kamrul
Islam, Md. Mahedi Hassan, M M Fazle Rabbi,
“Depression Detection System with Statistical Analysis
and Data Mining Approaches”, in 2021 International
Conference on ScienceandContemporarytechnologies,
December 2021.
[14] G H Suhas, L Suraj, J Varun, D V Veda, H S Jayanna,
“Machine Learning Approaches for Detecting Early
Stage Depression using Text”, in 20215thInternational
Conference on Electrical, Electronics, Communication,
Computer Technologies and Optimization Techniques,
February 2022.
[15] Aanchal Bhist, Shreya Vashisht,MuskanGupta,Ena Jain,
“Stress Prediction in Indian School Students using
Machine Learning”, in 2022 3rd International
ConferenceonIntelligentEngineeringandManagement
(ICIEM), August 2022.
[16] Akshada Kene, Shubhada Thakare,“Mental StressLevel
Prediction and Classification based on Machine

Learning”,in2021SmartTechnologies,Communication
and Robotics (STCR), November 2021.
[17] Somnath Sinha, Sriram R, “An Educational based
Intelligent Student Stress PredictionusingML”,in2022
3rd International Conference for EmergingTechnology
(INCET), July 2022.
[18] Anika Kapoor, Shivani Goel, “Prediction of Anxiety
DisordersusingMachineLearningTechniques”,in2022
IEEE Bombay Section Signature Conference (IBSSC),
February 2023.
[19] Ahnaf Atef Choudhury, Md. Rezwan Hassan Khan,
Nabuat Zaman Nahim, Sadid Rafsun Talon, Samiul
Islam, Amitabha Chakrabarty,“PredictingDepressionin
Bangladeshi Undergraduates using MachineLearning”,
in 2019 IEEE Region 10 Symposium (TENSYMP),
January 2020.
[20] DASS-21 Lovibond PF, Lovibond SH. Manual for the
Depression Anxiety & Stress Scales. 2nd ed. Sydney,
Australia: Psychology Foundation; 1995.

Mental Illness Prediction using Machine Learning Algorithms

Recommended

More Related Content

Similar to Mental Illness Prediction using Machine Learning Algorithms (20)

More from IRJET Journal (20)

Recently uploaded (20)

Mental Illness Prediction using Machine Learning Algorithms