Predicting Student Academic Performance Using Explainable Machine Learning: Implications for Institutional Decision-Making

Abstract

Educational institutions increasingly require timely and accountable evidence to identify students

who may benefit from academic support. This paper develops an explainable machine learning

framework for predicting academic performance and translating individual predictions into

institutionally defensible decisions. The proposed pipeline integrates longitudinal academic

records, attendance, learning management system activity, and contextual variables; performs

temporal validation; compares logistic regression, random forests, and gradient boosting; and

explains predictions using global feature importance, permutation analysis, and SHAP-style

additive attributions. It specifies calibration, subgroup fairness, intervention capacity, and

privacy safeguards as first-class evaluation criteria. To demonstrate reporting procedures without

claiming access to student records, the paper includes a clearly labeled synthetic numerical

illustration. In that illustration, gradient boosting achieves an accuracy of 0.84 and an area under

the ROC curve of 0.90, but these figures are not empirical findings and cannot substantiate

October–December 2023 www.ijhit.info 272 | PageInternational Journal of Humanities and Information Technology (IJHIT)

e-ISSN: 2456 –1142, Volume 5, Issue 4, (October–December 2023), ijhit.info

institutional effectiveness. The main contribution is a reproducible decision-support design that

connects predictive performance with human review, proportional intervention, and post-

deployment monitoring. The study identifies conditions under which explainable prediction may

inform advising, resource allocation, and academic policy while avoiding automation of high-

stakes judgments.