An Explainable Machine Learning Framework for Predicting Student Academic Outcomes in Higher Education

Abstract

The increasing use of artificial intelligence (AI) in higher education has opened up new possibilities for enhancing student achievement through data-driven decision making and predictive analytics. Higher education institutions can identify students who are at risk of academic failure or dropout and undertake timely interventions to increase retention and graduation rates by using early prediction of student academic outcomes. However, many machine learning models employed for student outcome prediction operate as black-box systems, limiting their transparency and reducing stakeholders' confidence in their predictions. This study proposes an explainable machine learning framework for predicting student academic outcomes in higher education using the Predict Students' Dropout and Academic Success dataset obtained from the UCI Machine Learning Repository. Five supervised machine learning algorithms, Logistic Regression, Decision Tree, Random Forest, XGBoost, CatBoost, and LightGBM were developed and evaluated for predicting three academic outcome categories: Dropout, Enrolled, and Graduate. The proposed framework provides an interpretable and effective decision-support tool that can assist higher education institutions in identifying at-risk students and implementing evidence-based academic interventions.