AI-Driven Weighted Ensemble Framework for Assessing and Predicting Maladaptive Behaviors in University Students

Abstract

The proposed research hypothesizes about an AI-based method of evaluating and forecasting maladaptive behaviors in computer science students. Validated psychometric scales were used to collect data about mental health, stress, sleep patterns, suicidal tendencies, and academic performance (DASS-21, ISI, and SBQ-R). Following preprocessing and feature encoding of data, various deep learning models DNN, CNN, LSTM, and BiLSTM were created to achieve nonlinear and temporal features. In general mental health risk prediction, a new weighted ensemble (WVE-RGS) combined the power of a Random Forest, Gradient Boosting, and SVM classifiers with a maximum accuracy of 97.6 and ROC-AUC of 99.8. The explainable AI approaches (SHAP and LIME) were very transparent because they gave the contribution feature-level, i.e. stress, depression and anxiety indicators. The framework facilitates the detection of vulnerable students early in life and encourages evidence-based, ethical, and collaborative mental health interventions. This study shows that AI-enabled learning communities can support the well-being of students, institutional stay, and responsible technology use in higher education by adhering to the principles of open innovation.