PREDICTING STUDENT ACADEMIC PERFORMANCE USING MACHINE LEARNING ALGORITHMS IN HIGHER EDUCATION

This comprehensive study constructs an advanced predictive analytics framework leveraging machine learning algorithms to forecast undergraduate academic performance within higher education institutions. Utilizing empirical student data from a technical institute, including Learning Management System (LMS) interaction logs and historical semester metrics, we implemented and evaluated four robust supervised classification architectures. Data preprocessing strictly utilized SMOTE techniques to systematically resolve extreme class imbalances. Experimental results demonstrated that ensemble methods, specifically XGBoost and Random Forest, significantly outperform baseline models, achieving a peak predictive accuracy of 92.4%. Mapped feature hierarchies reveal that early assignment submission latency and historical grade points act as critical early predictors for data-driven student intervention frameworks.