OptDCE: An Optimal and Diverse Classifier Ensemble for Imbalanced Datasets

Abstract

Machine learning has evolved dramatically in recent years and plays a very important role to ease the day-to-day activities. Classification is one of the major tasks in machine learning. It is concerned with the categorization of the data in various applications such as software fault detection, credit scoring systems and medical applications. Many of these applications suffer from the problem of Imbalanced data classification wherein one class consists of a large number of samples while samples representing another class are very less in number. The skewed nature of data results in the imprecise classification of the data which may be very harmful in some disciplines like medical applications. To highlight the class imbalance issue, this work presents the impact of the increased degree of class imbalance on the classification performance of various datasets. Moreover, we present the classification approach that integrates the data level technique with a diverse classifier ensemble (CE). The experimental results show significant improvements in the classification performance of imbalanced datasets.