Enhancing Classification Performance Through Hybrid Feature Subset Selection and Eadaboost

Abstract

Feature subset selection plays a crucial role in improving classification accuracy by identifying the minimal set of relevant attributes. In this work, we propose an integration framework that combines a hybrid feature subset selection technique with an effective AdaBoost (eAdaBoost) algorithm to enhance classification performance. Unlike traditional AdaBoost, the eAdaBoost variant adaptively reweights features, thereby reducing error rates and improving accuracy across diverse datasets. Experiments were conducted using multiple benchmark datasets from the UCI Machine Learning Repository, with classifiers including Decision Stump, C4.5, Random Forest, and NBTree. The proposed hybrid-eAdaBoost integration consistently outperformed existing approaches, achieving higher classification and prediction accuracy with reduced execution time. Comparative analysis further demonstrates that eAdaBoost yields lower or comparable error rates relative to standard AdaBoost across most datasets. These results highlight the robustness and efficiency of the proposed method, providing a promising direction for future developments in ensemble-based classification.