Published Paper Details:

Student performance prediction is a crucial research domain which analyses the student data aiming to anticipate academic outcomes, analyze student academic performance and to identify student at risk. In this research paper one of the efficient approaches called SHAP (SHapley Additive explanations) which is mathematically grounded with game theory is used to make feature selection and final predictions. SHAP provides both global explanations (which feature is important overall) and local explanations (why a prediction was made for a particular student). In this study, we employ different machine learning algorithm Decision Tree, Random Forest, Lasso Regression, and XGBoost with SHAP based analysis for feature selection. The work is carried out with k-fold validation technique, feature overlaps and residual plots. Paired t-test is also used to verify results statistically. Results showed that SHAP integrated with Random Forest importance based selection identified a more stable and interpretable feature subset, outperforming other combinations. The study concludes that integrating SHAP with traditional feature selection improves both predictive reliability and explainability.