Accurate and Explainable AI in Student Performance Prediction Using E-Learning Classification

The proposed research utilizes a 3-tier hybrid feature selection technique which connects Boruta Optimization to Ridge (L2) Regularization for eliminating noise and reducing redundancy during the selection process. This research implements a Multi-Tier Ensemble model that combines Random Forest, SVM and Gradient Boosting to overcome the limitations of conventional student performance prediction systems which emphasize accuracy at the expense of explainability and real-time adaptability together with optimized feature selection. A combination of Bayesian Optimization PSO with Random Search enables the optimization process to achieve improved efficiency along with robust resistance. Z-score normalization together with SMOTE techniques improve data quality by handling dataset outliers and distribution imbalances. The research includes Explainable AI (XAI) functionality which utilizes LIME for maintaining clear understandability. The empirical model surpassed previous models by reaching 98.74% accuracy and 98.76% precision while obtaining 99.88% AUC. The upcoming research will examine real-time system adaptability as well as alternative XAI methods and multi-source data integration to boost the predictive capability and real-world practicality of educational analytics systems.