LINEAR REGRESSIVE LAPLACIAN KERNELIZED DEEP SHAP ALGORITHM FOR PREDICTING STUDENT ACADEMIC PERFORMANCE AND ANALYZING INFLUENCING FACTORS

Abstract

             Student academic performance is considerable results of a student's learning, measured by grades, test scores, assignments, and overall achievements. It reflects the student's kind of subjects and their abilities. Predicting student academic performance involves forecasting future grades or achievements using historical data, namely, previous grades, attendance, study habits, and personal background. Academic performance is partial by different factors are such as, including individual effort, teaching methods, and environmental conditions. Although several techniques have been used for prediction, achieving accurate results remains challenging. To address these issues, this study introduces a novel Linear Regressive Laplacian Kernelized Deep SHAP (LRLKDS) framework that integrates regression-based feature conditioning with kernel-weighted explainability. The model first employs a Linear Regressive Deep Multiplayer Classifier (LR-MLP) to impute missing values and stabilize learning, followed by a Laplacian Kernelized Piecewise SHAP (LK-P-SHAP) module to derive interpretable feature attributions. Experiments conducted on the Kaggle Student Performance dataset demonstrate that the proposed method achieves superior classification accuracy (97.25%) and improved interpretability over existing approaches such as LSTM and MSPP. Statistical analysis confirms that the observed performance gains are significant at the 95% confidence level. The LRLKDS framework thus offers an efficient, interpretable, and statistically validated solution for educational data analytics.