EDUSTACK-MH: An Intelligent Ensemble-Based Model for University Student Depression Screening

The increasing mental health crisis amongst university students has made early detection of depression a pressing concern. Despite being clinically validated, the current diagnostic tools are frequently subjective, time-consuming, and have limited scalability. This study presents EDUSTACK-MH, a novel ensemble-based machine learning framework that uses structured data from the DASS-21 scale and sociodemographic parameters to identify students' depression severity. Data were collected from 1,120 university students in Dehradun, India, and preprocessed to build predictive models capable of identifying five levels of depression severity-ranging from normal to extremely severe. A comparative analysis of baseline classifiers (KNN, SVC, Random Forest, Gradient Boosting, Logistic Regression) and ensemble approaches (Optimized Voting and Stacking) demonstrated that the Optimized Stacking model outperformed all others, achieving 98.23% accuracy with near-perfect classification performance. The findings demonstrate that early mental health screening and intervention techniques in educational environments can be significantly enhanced by combining intelligent ensemble learning techniques integrated with validated psychological tests. By providing a dependable, scalable, and decent method for determining the full extent of depression, this study addresses a significant gap and has been specifically developed to meet the needs of Indian college students.