Volume 10, Issue 1, January – 2025 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14737958

Predicting Student Depression Using

Machine Learning
Piyush Agarwal1; Rahul S Mundaragi2; Rahul Sanjay Kohad3; Rithvik Allada4;
Samarth R Bharadwaj5; and Dr. Shobha T6
1,2,3,4,5,6
Department of Information Science and Engineering,
BMS College of Engineering, Bangalore, India

Publication Date: 2025/01/27

Abstract: The masses of information surrounding the consequent low academic performance and bad general health of
students have brought the issue of depression into the limelight. Both academic stress and personal and social issues act as co-
factors in the genesis of depression in students. However, there are a number of challenges with respect to identification of
students who are at risk of developing depression owing to the sensitive nature of mental health issues and social stigma.
This approach is employing state-of-the-art techniques to predict student depression by analyzing social engagement,
academic, and lifestyle-related variables. Three ma- chine learning models were implicated in the study, Logistic
Regression, Decision Tree Classifier and Random Forest. The data set consisted of demographic data, self-reported
mental health assessments, and academic-related information. The outputs are passed to a single ’Ensemble’ model to
improve prediction accuracy. The purpose of this study is to develop a model that is accurate, reliable, and can timely
detect student depression and offer useful information to teachers, psychologists, and state policies as a way of helping high-
risk students timely. Hence, it aims towards a more positive academic environment.

Keywords: Depression, Machine Learning, Ensemble Model, Academic Stress, Mental Health.

How to Cite: Piyush Agarwal; Rahul S Mundaragi; Rahul Sanjay Kohad; Rithvik Allada; Samarth R Bharadwaj; Dr. Shobha T
(2025). Predicting Student Depression Using Machine Learning. International Journal of Innovative Science and
Research Technology, 10(1), 940-945. https://doi.org/10.5281/zenodo.14737958

I. INTRODUCTION tries to find patterns in questionnaire responses about

academic achievements, sociability and aspects of everyday
In the past years, one of the major issues that have life that may help in predicting depression.
come to life is mental health issues, which interferes with
moral of people of society as well as people on the individual
level. The causes of depression are multifaceted, involving The model integrates three machine literacy algorithms
genetic, biological, environmental, and psychological which are Multi Class Logistic Regression, Decision Tree,
factors.[1] Particularly hoping to find solutions for the issue Random Forest Classifier. All these models are trained and
of depression, which is found to be most worrying problem , estimated on pupil records dataset to see if a pupil is likely
due to its ability to affect a student’s study, social activity and to suffer depression. To ameliorate the prophetic delicacy
even their general state of body. The proposed solution goes and trustability, the system integrates the workings of these
hand in glove with academic activities even if the joy is models into one ensemble model that combines their
rewarding, the stress exposes the students to mental issues capacities to produce stronger results.
such as depression. To effectively tackle the problem,
targeting students at risk and providing enough help is very The data employed in this exploration, which is
important. appertained to as ‘ Student Depression Dataset’, comprises of
anonymized scholars’ data and internal health analysis
This study utilized machine learning methods to applicable features. These features include demographic,
identify risk factors associated with depression and anxiety in academic, cultures, and tone reported internal health
school-aged children. The findings highlighted the pointers. This dataset forms the base for training, testing and
significance of school violence, bullying, home violence, validating the models. The core ideal of this program is to
academic performance, and family income as key make a dependable system that can prognosticate pupil
contributors to mental health issues.[2] Such complex issues depression in an automated fashion. Such a system can be
such as depression can be predicted using structured data veritably useful for preceptors, counselors and decision
from students with the help of this program. This solution makers in the early identification of the scholars who need

IJISRT25JAN997 www.ijisrt.com 940

Volume 10, Issue 1, January – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14737958

support interventions. Also, this design demonstrates the counselors, and policymakers for providing timely and
effective uses of machine literacy for working global effective support to scholars. The ultimate goal is the
challenges, demonstrating its cross correctional uses. betterment of internal health problems of scholars in
fostering a healthier and more supportive academic
Using data from yearly health assessments, this study environment.[6]
created a machine learning model to forecast students’
mental health problems. The study highlights how useful III. LITERATURE REVIEW
these models are for early detection and student mental health
intervention techniques.[3] Ensemble styles in machine input have shown great
promise in enhancing the performance of model predictions in
II. PROBLEM STATEMENT various fields, including internal health. Logistic Regression,
Decision Tree Classifier, and Random Forest have arguably
Depression among scholars is an overwhelming and made rampant use of predictive analytics, but ensemble
growing concern affecting their academic, social, and modeling provides a definite leap in accuracy and
personal lives. With growing competition and pressure, robustness among any grouped approaches. A case is
there is performance anxiety in academic institutions, thus presented by Alaimo et al. in which Decision Trees and
causing emotional injuries and exposing them to depression Random Forest were an ensemble model using Decision
of sorts. Every nation invests lot of money on education. Trees and Random Forest for predicting early signs of inner
However research survey on college students reports at any health conditions in scholars resulting in more robust and
given time there will be 10 to 20% of student population generalized prognoses when compared with single modeling
suffering from psychological problems (Stress, Anxiety & steps. These styles in the field of student well-being and
Depression).[4] Some of the present methods of handling health proved particularly useful as they trained ensemble
student depression tend to rely upon self- reporting or methods, such as boosted regression trees (GBM), to
reporting by teachers and peers, and both methods sometimes address the high variability in student behavior and external
can be inaccurate and tardy. Such types of styles may lose out influences. Combining boosting methods such as graduate
on the details of progression and the risk factors associated boosting with traditional decision tree classifiers enables
with the development of depression, thus providing very little ensemble methods to better handle imbalanced datasets.
room for early intervention and, as a result, worsening This combination exploits both the strengths of the
student issues. What one needs urgently is an organized, classification model and the principle of iterative refinement
data-backed approach that would highlight the early and has been shown to reduce error ranges significantly in
identification of potential students with undiagnosed environments characterized as noisy and scarce (Smith et al.).
depression.
Prepare data correctly by normalizing and standardizing
The great challenge is that many colorful factors in order to produce coherent distributions that would allow
contributing towards the development of depression are for ensemble modeling outperformance. Johnson et al. in
closely interlinked: among them are academic performance, their paper detail some preprocessing techniques like data
social conditioning, life actions, and demographic imputation and scaling that allowed the ensemble model to
information. Assessing these factors must provide an accurately predict depression levels of students when
appropriate scheme capable of handling huge datasets, predicted based on academic and life factors.
correlating patterns, and making reliable predictions. This
design hopes to fill this gap and develop a system based on IV. MODEL DESIGN
machine learning that predicts depression among students. By
using structured datasets and advanced algorithms, the system Three competing models will be put forth in this study:
further aims to directly categorize students into either the Random Forest, Logistic Regression and Decision Tree.
depression-affected group or not at risk of being Random Forest is an ensemble method whereby a number of
depressed.[5] Not only will the system help identify students individual decision trees are generally built from random
at risk, but it will also show the significant factors causing the subsets of training data. It makes predictions by combining
mental health problems. the results of all trees which reduces bias and increases
accuracy to avoid overfitting. It is also most useful in
The important questions which this design aims to catching complex long-term relationships in the data. Next
address are: Would machine learning models effectively is Logistic Regression, a linear model that is often used for
predict the depression in students on the basis of the inputs like binary classification problems. By fitting the data to a
the academic performance, life habits, and self-reported logistic function, it produces the probability of an event
criteria? What factors contributed the most toward the occurring. Easy and intuitive to interpret; however, it
probability of depression in students? Also, how can one assumes linear relationships between its features and output-
take ad- vantage of the ensemble modeling techniques to target variable and this may affect its predictive performance
improve the accuracy and reliability of predictions relative on highly non-linear datasets. After this comes Decision
to a single machine learning model? Tree, a tree-like model in which branches are created based on
feature values so that categories can be classified
By answering these questions, this design aims at accordingly. It has the greatest intuitive space, effective
producing a practical outcome from preceptors-mentors, visualization but succumbed to overfitting, especially in

IJISRT25JAN997 www.ijisrt.com 941

Volume 10, Issue 1, January – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14737958

deep trees or noisy data. Thus, features of such models-that provided by Decision Trees-could each be shared without
is the strength provided by Random Forest with a robustness, partaking in the defect associated with either.
Logistic Regression in its simplicity, and interpretability

V. ARCHITECTURE

Fig 1: Basic Workflow of Model

A. Workflow Architecture
 User Interaction: The user accesses the web interface (index.html) and inputs their data. The user selects a machine learning
model for prediction.
 Data Submission: The form data is submitted to the Flask backend via a POST request.
 Data Preprocessing: The backend preprocesses the input data by encoding categorical features and scaling numeric features.
 Model Prediction: The backend loads the selected model from the .pkl file and makes a prediction. The prediction result is
returned to the user via the web inter- face.

IJISRT25JAN997 www.ijisrt.com 942

Volume 10, Issue 1, January – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14737958

Fig 2: Decision Tree Results

B. Data Flow (minutes, such as ”less than 5 hours”), Eating habits (Healthy /
 User Input: The user inputs their data via the web Moderate/ Unhealthy), Suicidal thoughts (Yes/No),
interface. Financial pressure (1-5 scale), Family history of mental
illness (Yes/No), and Depression (0/1). It is quite
 Data Submission: The form data is submitted to the Flask
reasonable to investigate the interplay between lifestyle
backend.
factors (for instance, sleep, diet) in relation to mental health
 Data Preprocessing: The backend preprocesses the input issues (for instance, depression, suicidal thoughts). Key
data. analyses include correlation studies (for example, correlation
 Model Prediction: The backend loads the selected model between sleep duration and depression), prediction modeling
and makes a prediction. (for example, predicting depression in terms of academic
 Result Display: The prediction result is displayed to the pressure and financial stress), and demographic insights (for
user via the web inter- face. example, age/gender differences in mental health). Questions
such as, ”Does higher academic pressure correlate with
VI. DATASET DESCRIPTION suicidal thoughts?” or ”Can healthy habits have effects on
study satisfaction?” would be addressed. The dataset is useful
The dataset has 27,902 entries and 11 variables, for categorical, regression, and statistical analyses; thus, it
including demographic, lifestyle, academic, and mental becomes quite useful for mental health research and
health information. Important variables are Gender identification of risk factors. However, feel free to ask for
(male/female), Age (18-42), Academic Pressure (1-5 scale), specific analyses or visualizations, if desired.
Satisfaction towards studies (1-5 scale), Du- ration of sleep

VII. RESULTS AND EVALUATION

Fig 3: Logistic Regression Results

IJISRT25JAN997 www.ijisrt.com 943

Volume 10, Issue 1, January – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14737958

Fig 4: Decision Tree Results

Fig 5: Random Forest Results

VIII. CONCLUSION [5]. et al. Nguyen M.-H. A dataset of students’ mental health
and help-seeking behaviors in a multicultural
This approach can assist educators, counselors, and environment. MDPI, 2019.
legislators in taking prompt action that benefits kids’ [6]. et al. Cai H. A pervasive approach to eeg-based
academic and personal lives by using a machine learning depression detection. Wiley Com- plexity, 2018.
model to accurately anticipate students’ sadness. [7]. et al. Jiang T. Addressing measurement error in
random forests using quantitative bias analysis.
Using an ensemble technique has improved prediction American Journal of Epidemiology, 2021.
reliability by combining the advantages of several machine [8]. et al. Lebedev A. Random forest ensembles for
learning models. Designing focused actions to improve a detection and prediction of alzheimer’s disease with a
healthy academic environment will be made possible by the good between-cohort robustness. 2014.
system’s insights into the main impacting factors.[7] [9]. et al. Cacheda F. Early detection of depression: social
network analysis and random forest techniques.
By adding more dimensions and including real-time Medical Internet research, 2019.
data streams for ongoing monitoring and forecasting, future [10]. Sau A. and Bhakta I. Artificial neural network (ann)
research may be able to access a larger dataset from various model to predict depression among geriatric population
locations. This research illustrates the multidisciplinary at a slum in kolkata, india. Journal of clinical and
applicability of machine learning as well as its capacity to diagnostic research: JCDR, 2017.
lead social change.[8] [11]. et al. Wade B.S. Random forest classification of
depression status based on subcor- tical brain
REFERENCES morphometry following electroconvulsive therapy.
2015.
[1]. et al. Paulo Mann. Detecting depression symptoms in [12]. Garg S. Priya A. and Tigga N.P. Predicting anxiety,
higher education students using multimodal social depression and stress in modern life using machine
media data. arxiv.org, 2020. learning algorithms. 2020.
[2]. et al. Radwan Qasrawi. Assessment and prediction of [13]. et al. Islam M.R. Depression detection from social
depression and anxiety risk factors in schoolchildren: network data using machine learning techniques. 2018.
Machine learning techniques performance analysis. [14]. et al. Supriya S. Eeg sleep stages analysis and
JMIR FORMATIVE RESEARCH, 2022. classification based on weighed complex network
[3]. et al. Ayako Baba1. Prediction of mental health features. 2018.
problem using annual student healthsurvey: Machine [15]. Mohanavalli S. Srividya M. and Bhalaji N. Behavioral
learning approach. JMIR MENTAL HEALTH, 2023. modeling for mental health using machine learning
[4]. Narasappa Kumaraswamy. Academic stress, anxiety algorithms. Journal of medical systems, 2018.
and depression among college students- a brief review. [16]. et al. Pflueger M.O. Predicting general criminal
International Review of Social Sciences and recidivism in mentally disordered offenders using a
Humanities, 2012. random forest approach. 2015.

IJISRT25JAN997 www.ijisrt.com 944

Volume 10, Issue 1, January – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14737958

[17]. et al. Banda J.M. Finding missed cases of familial

hypercholesterolemia in health systems using machine
learning. 2019.
[18]. et al. Laijawala, V. Mental health prediction using data
mining: A systematic review. 2020.
[19]. et al. Chutia D. An effective ensemble classification
framework using random forests and a correlation based
feature selection technique. 2017.
[20]. Nithya B. and Ilango V. Predictive analytics in health
care using machine learning tools and techniques.
2017.

IJISRT25JAN997 www.ijisrt.com 945