Predicting Depression in College Students Using DynamicWeighted Ensemble Learning: An Explainable Artificial Intelligence Approach

Depression among college students is a global public health challenge. Traditional screening methods often suffer from poor timeliness and ‎low sensitivity. To address this, we constructed a Dynamic Weighted Ensemble Model (DWEM) that integrates five algorithms: CatBoost, ‎XGBoost, LightGBM, Random Forest, and ExtraTrees, with ensemble weights optimized using the Optuna framework [1]. Employing ‎stratified 5-fold cross-validation, the model achieved an accuracy of 94.96% ± 0.44% and an AUC of 98.95% ± 0.12%, demonstrating ‎exceptional discriminatory performance and stability. Furthermore, explainability analysis via the SHAP framework not only identified core ‎risk factors such as academic pressure and sleep problems but also facilitated the development of a tiered intervention strategy based on ‎predicted probabilities and feature contributions. Our study confirms that combining advanced ensemble learning with Explainable AI (XAI) ‎can provide a powerful tool for shifting college mental health management from a “passive response” to an “active defense” paradigm, hold‎ing significant potential for clinical translation‎.