A Machine Learning Approach to Detect Depression from Student Mental Health Surveys

Abstract

The increasing instances of depression among students have become a significant global mental health con-cern, creating an urgent need for effective early detection methods. This study addresses the growing con-cern of depression among students by proposing a machine learning based system for predicting depression risk levels using structured questionnaire data. The dataset consisted of psychological and behavioral indica-tors, including interest loss, feelings of sadness, sleep issues, low energy, appetite issues, low self-worth, con-centration difficulties, movement changes, self-harm thoughts, scholarship status, etc. After data collection, comprehensive preprocessing and feature selection techniques were applied to improve data quality, reduce redundancy, and enhance predictive performance. Random Forest and XGBoost classifiers were implemented to classify students into different levels of depression severity. Experimental results showed that both models performed effectively in identifying at-risk students, with XGBoost achieving the highest accuracy of 88.64% on unseen test data. Feature analysis revealed that sleep issues, feelings of sadness, low energy, and self-harm thoughts were among the most influential factors affecting prediction outcomes. The proposed system demonstrates the potential of machine learning techniques for early, scalable, and data-driven mental health assessment. It can support educational institutions, counselors, and healthcare professionals in identifying students who may require timely psychological intervention, thereby contributing to improved mental well-being and preventive mental healthcare in academic environments.