ElimRidge-HFM: A Mathematical Integration of Feature Selection, Boosting, and Hyperparameter Search for Enhanced Predictive Performance in E-Learning

Abstract

The rising demand for e-learning platforms requires reliable predictive models to improve learning results. This study establishes ElimRidge-Hybrid Fusion Model (HFM) as a new predictive framework for e-learning data sets that solves the problems of feature selection, model overfitting, and potential dataset complexity. Thus, the objective of the study is as follows: Enhance the predictive precision and interpretability of the model by applying feature selection techniques using Ridge Regularization (L2) in conjugation with Recursive Feature Elimination (RFE) and using a Hybrid Fusion Model (HFM), which is the combination of AdaBoost and CatBoost. For the hyperparameter tuning, Gaussian Process-Enhanced Random Search (GP-RS) is used to explore the hyperparameters of the space efficiently with the least number of random trials. For data clearing this framework uses Synthetic Minority Over-sampling Technique (SMOTE) for class balancing, Interquartile Range (IQR) for outlier removal and, Z-score normalization for normalization. Experimental results demonstrate that ElimRidge-HFM significantly improves predictive performance, achieving 97% accuracy, 97% precision, and 97% F1-score. Statistical validation (p < 0.05) confirms its superiority over traditional models, effectively handling imbalanced and noisy e-learning datasets. These findings highlight its potential for enhancing personalized learning strategies and optimizing student engagement. These results further strengthen the claims that ElimRidge-HFM has great ability to handle imbalanced and noisy data sets due to its scalability. This research contributes to the development of e-learning techniques by providing opportunities to introduce decisions and/or strategies when the students’ performance is low or during cases of inequality in students’ learning opportunities.