Clustering-based subset ensemble learning method for imbalanced data

Abstract

In recent research, classification involving imbalanced datasets has received considerable attention. Most classification algorithms tend to predict that most of the incoming data belongs to the majority class, resulting in the poor classification performance in minority class instances, which are usually of much more interest. In this paper we propose a clustering-based subset ensemble learning method for handling class imbalanced problem. In the proposed approach, first, new balanced training datasets are produced using clustering-based under-sampling, then, further classification of new training sets are performed by applying four algorithms: Decision Tree, Naïve Bayes, KNN and SVM, as the base algorithms in combined-bagging. An experimental analysis is carried out over a wide range of highly imbalanced data sets. The results obtained show that our method can improve imbalance classification performance of rare and normal classes stably and effectively.