BOOSTING STUDENT PERFORMANCE PREDICTION IN E-LEARNING: A HYBRID FEATURE SELECTION AND MULTI-TIER ENSEMBLE MODELLING FRAMEWORK WITH FEDERATED LEARNING

This research introduces a novel and advanced methodology for predictive modeling using federated learning, addressing critical challenges such as data privacy, class imbalance, and model performance. Unlike traditional centralized approaches, our work ensures data privacy through federated learning, enabling high-performance models without exposing sensitive data. The novelty of our approach lies in the integration of advanced preprocessing techniques, such as the Synthetic Minority Oversampling Technique (SMOTE) for class imbalance, hybrid feature selection by the combination of Boruta algorithm and L2 regularization's (Boruta-L2) for robust feature selection, and a 3-tier ensemble model with cutting-edge hyperparameter tuning techniques, including Bayesian Optimization, Random Search, and Particle Swarm Optimization (PSO). As a result, our global model achieves an accuracy of 98.90%, significantly outperforming previous methodologies. The advancements in our work are highlighted by the superior model performance, scalability, and privacy preservation, making it a significant contribution to federated learning. This research provides a comprehensive, efficient, and privacy-preserving solution for distributed predictive tasks, setting a new benchmark in machine learning applications across various domains.