Early Detection of Depressive Tendencies Among Filipino College Students Using Machine Learning: A Demographic-Based Predictive Model

The growing incidence of depression among college students remains a significant public health concern, particularly in low- and middle-income countries where access to mental health services is limited. This study explores the viability of machine learning (ML) techniques for early detection of depressive tendencies using non-clinical, demographic, and self-report data collected from 1,585 students across 38 higher education institutions in Quezon Province, Philippines. Employing a quantitative cross-sectional design, the study trained and evaluated six supervised ML models including Logistic Regression, Random Forest, Gradient Boosting, LightGBM, Artificial Neural Networks, and Discriminant Analysis against performance metrics including accuracy, precision, recall, F1-score, and AUC using Python and deployed results via a Streamlit interface for interactive visualization. LightGBM emerged as the top-performing model with 94% accuracy and a 0.97 AUC score. Feature importance analysis revealed that living arrangement, family type, and academic program were key predictors of depressive symptoms. The findings demonstrate that scalable, non-invasive, and context-aware predictive tools can enhance institutional mental health interventions. The study also highlights the integration of psychosocial theory, specifically the Diathesis-Stress Model, into ML frameworks for more ethically sound and culturally relevant detection systems.