A HYBRID XGBOOST-LSTM FRAMEWORK FOR SCALABLE ASSESSMENT OF TIME MANAGEMENT COMPETENCE IN HIGHER EDUCATION: SHAP-DRIVEN INSIGHTS FROM WEST BENGAL COHORT

Effective time management is vital for undergraduate students to succeed in demanding academic environments, yet scalable assessment tools remain limited. This study introduces a hybrid XGBoost-LSTM framework, integrated with a Python Flask-based web application, to evaluate time management competence among 313 undergraduate students at a college in West Bengal, India. A PCA validated 10-question quiz, derived from a 31-item survey, demonstrated high reliability with Cronbach’s Alpha equal to 0.87. The XGBoost model classified students into Poor, Average, or Good categories with an accuracy of 90% and an F1-score of 0.89, while a RandomForestRegressor achieved an RMSE of 0.21, improving 75.65% over the baseline. SHAP-based analysis identified delaying tasks and scheduling as key predictors. A significant gender difference was found (p=0.013), but no residence differences (p=0.43). A simulated LSTM model was implemented as proof-of-concept for future longitudinal analysis, with an RMSE of 0.21. The Flask application provides real-time categorization and feedback, offering a scalable tool for identifying students needing support. Future work includes longitudinal data collection and cloud-based deployment to enhance regional educational insights.