Predicting Student Academic Success Using Comparative Machine Learning and Explainable Artificial Intelligence

Abstract

Student academic success prediction is a significant educational data mining task because it allows early intervention, enhances retention, and assists with academic planning. This paper investigates whether demographic, academic history, and socioeconomic factors can predict student outcomes in the Open University Learning Analytics Dataset (OULAD). The dataset (studentInfo.csv) consists of 32,593 student records with 12 variables. The final analytical dataset comprised 21,562 observations after excluding withdrawn students and recording the result as a binary target. This analysis involved a descriptive exploration and analysis of socioeconomic deprivation using the Index of Multiple Deprivation (IMD) band. A chi-square test showed that the IMD band and the final academic outcome are statistically associated (χ² = 472.42, p < 0.001), indicating that performance depends on socioeconomic background. Three supervised machine learning models, Logistic Regression, Decision Tree, and Random Forest, were developed and compared. Logistic Regression achieved the highest discrimination performance (ROC-AUC = 0.672), while Random Forest achieved the highest recall (0.794) and F1-score (0.748). To increase interpretability, feature importance, permutation importance and SHAP analyses were used. Findings have shown that the most significant predictors of academic success are prior educational experience, frequency of prior attempts, credits completed, age, and selected socioeconomic variables. The implications of these results are that interpretable machine learning offers a useful structure to student risk identification and that structural and academic preparedness factors remain relevant in defining student success.

Keywords-Educational Data Mining, Explainable Artifitial Intelligence, Machine Learning, Predictive Analytics, SHAP, Student Performance Prediction

DOI: 10.7176/JEP/17-5-05

Publication date: May 30^th 2026