Machine Learning-Based Prediction of Academic Performance from Mental Health and Behavioural Features: A CBT-Integrated Approach for Student Mental Health Assessment

Student mental health significantly influences academic performance, yet early identification of at-risk individuals remains challenging. This study develops machine learning models to predict mental health status using psychological and behavioural features. A dataset of 500 university students was analysed with nine features, including stress level, anxiety score, depression score, sleep hours, physical activity, and academic metrics. Three algorithms were evaluated: Logistic Regression, Support Vector Machines (SVM), and Random Forest. Results demonstrate that Random Forest achieves 98% test accuracy (F1-Score: 0.9779, ROC-AUC: 1.0000) with 5-fold cross-validation, F1: 0.9736 ± 0.0106. SVM achieves 86% accuracy (F1-Score: 0.8417, ROC-AUC: 0.9756), while Logistic Regression achieves 81% accuracy (F1-Score: 0.8022). Depression score (42.22%), anxiety score (24.16%), and stress level (19.05%) emerged as dominant predictors, accounting for 85.5% of model decisions. Cross-validation analysis confirms robust generalization of the Random Forest model. The findings support the integration of psychological screening within academic institutions to identify students requiring cognitive behavioural therapy interventions. This research demonstrates the feasibility of implementing ML-based mental health prediction systems in educational settings as early warning mechanisms.