Enhanced Bagging (eBagging): A Novel Approach for Ensemble Learning

Abstract

Bagging is one of the well-known ensemble learning methods, which combines several classifiers trained on different subsamples of the dataset. However, a drawback of bagging is its random selection, where the classification performance depends on chance to choose a suitable subset of training objects. This paper proposes a novel modified version of bagging, named enhanced Bagging (eBagging), which uses a new mechanism (error-based bootstrapping) when constructing training sets in order to cope with this problem. In the experimental setting, the proposed eBagging technique was tested on 33 well-known benchmark datasets and compared with both bagging, random forest and boosting techniques using well-known classification algorithms: Support Vector Machines (SVM), decision trees (C4.5), k-Nearest Neighbour (kNN) and Naive Bayes (NB). The results show that eBagging outperforms its counterparts by classifying the data points more accurately while reducing the training error