Artificial Intelligence for Student Performance Prediction: A Systematic Review for Explainable Learning Analytics

Abstract

This research aims to systematically review the existing literature on AI-based student performance prediction studies in higher education to identify the seven structural gaps and suggest the Explainable Learning Analytics Framework (ELAF) as an integrated eight-layer conceptual architecture to cover this persistent gap between the prediction accuracy of algorithms and the use of the algorithms in the educational domain. A Systematic Literature Review based on PRISMA 2020 guidelines with inclusion/exclusion criterion, quality assessment, and qualitative synthesis were used to select studies published since 2018 that used AI/machine learning in higher education and student performance prediction. The reviewed literature shows that the most popular algorithms are Random Forest, XGBoost, Decision Trees, Support Vector Machines and Neural Networks with no one algorithm being consistently superior across the institutional contexts. It depends on the quality of data, features selected, more than algorithm choice, on prediction performance. Seven structural gaps were identified: the limited generalizability across institutions, underutilization of the multi-dimensional data, lack of explainable AI as a design requirement, weak prediction-to-intervention connectivity, ethical and privacy deficits, the lack of continuous real-time monitoring and disconnection from institutional quality assurance. Keywords: educational data mining, learning analytics, explainable AI, higher education, systematic review, SHAP, LIME