ANALYZING E-LEARNING STUDENT PERFORMANCE AND FEATURE IMPACT USING XGBOOST AND SHAP

E-learning has become an essential part of modern education, providing flexibility and accessibility to students worldwide. However, understanding the factors that influence student success in online learning remains a challenge. In this study, we analyze student performance in e-learning courses using the XGBoost classifier and SHAP (SHapley Additive exPlanations) to identify key features that impact academic outcomes. The dataset, obtained from the UC Irvine Machine Learning Repository, contains student information related to personal background, family environment, and educational habits. We categorize student performance in three ways: predicting exact grades, classifying students as pass/fail, and determining whether they achieve good or not good grades. Our findings show that certain features significantly influence student performance. For pass/fail prediction, factors such as last semester’s GPA, expected graduation GPA, and parental education levels were among the most impactful. In contrast, for good/not good classification, features like course ID, having a job, and accommodation type played a major role. The prediction results show that the model performs best in pass/fail classification, achieving an accuracy of 97.43%, while good/not good classification reached 80.07%. This suggests that predicting broad categories like pass/fail is much easier than predicting exact grades. These findings highlight the importance of carefully selecting features when analyzing student performance. Future research could expand this study by using a larger and more diverse dataset, incorporating additional features such as student engagement metrics, online activity logs, and psychological factors.