International Journal of Advanced Research in Computer and Communication Engineering

Abstract: Accurately estimating student academic performance is central to educational planning and early intervention strategies. Performance outcomes are influenced by multiple academic, behavioral, and lifestyle factors, making predictive modeling an important tool for educators and institutions. This study proposes a machine learning framework integrated with explainable artificial intelligence (XAI) to predict students' performance scores using features such as study duration, previous academic results, extracurricular participation, sleep patterns, and practice of sample question papers. The dataset was preprocessed through imputation of missing values and exploratory visualization techniques, including histograms, kernel density plots and correlation heatmaps, to assess distributions and identify anomalies. A diverse set of regression models, including Linear Regression, Bayesian Ridge, Ridge, Lasso, Elastic Net, Decision Tree Regressor, Random Forest, XGBoost, and LightGBM, was evaluated using MSE, MAE, RMSE, R², PSNR, and SNR. Linear Regression emerged as the best-performing method, achieving an MSE of 4.06, MAE of 1.60, RMSE of 2.01, R² of 0.98, and the highest PSNR and SNR values. To improve interpretability, SHAP and LIME techniques were applied to identify both global and local feature influences. The findings demonstrate that interpretable models supported by XAI can provide accurate predictions while enhancing transparency, thereby offering meaningful insights for educational research and policy formulation.

Keywords: Machine Learning, XAI, SHAP, LIME.

Downloads: | DOI: 10.17148/IJARCCE.2025.1411119