International Journal of Science, Engineering and TechnologyAuthors: Dipti D. Mehare
Abstract: Early Warning Systems (EWS) in higher education use data to identify students at risk of poor academic outcomes so institutions can deliver timely interventions. This paper presents the design, implementation, and evaluation of an AI-powered Early Warning System that integrates institutional records, learning management system (LMS) activity, and assessment data to predict at-risk students and produce actionable, explainable alerts for instructors and student support teams. The system architecture combines feature engineering, an ensemble predictive model (XGBoost), and model-agnostic explainability (SHAP) to provide both accurate predictions and interpretable risk drivers. An experimental evaluation on a multi-year dataset of undergraduate records demonstrates the feasibility of the approach and highlights operational considerations, privacy/ethical issues, and best practices for deployment. The work contributes a reproducible design blueprint and practical lessons for institutions seeking to implement AI-driven EWS.
