Enhancing Educational Quality with Explainable AI: Interpretable Prediction of Student Success

Abstract

Ensuring academic success and educational equity has become a critical priority for higher education institutions worldwide. This study investigates the determinants of academic success through interpretable machine learning and explainable artificial intelligence techniques. A dataset comprising 80,000 students was utilized to develop multiple regression models for GPA prediction, followed by SHAP dependence plots and Individual Conditional Expectation (ICE) plots to examine both global and instance-level contributions of behavioral, psychological, and environmental factors. Local interpretability was further explored through SHAP waterfall plots for selected students. Results indicate that study hours, motivation level, stress, sleep duration, study environment, and access to tutoring exert particularly strong influences on academic outcomes. The findings underscore that educational quality hinges on holistic student well-being beyond curriculum design alone. By delivering transparent, evidence-based explanations at both aggregate and individual levels, this research advances United Nations Sustainable Development Goal 4 (Quality Education) and offers higher education institutions actionable, data-driven guidance for equitable, student-centered enhancement strategies.