A Multi-Objective Co-Operative Co-Evolutionary Method for Classification with Imbalanced Data

In this paper, the authors propose a co-operative co-evolutionary ensemble algorithm (named IBMCCA) for imbalanced data classification. IBMCCA resolves the imbalance issue by simultaneously optimizing multiple objectives, including maximizing classification accuracy for both majority and minority classes and minimizing the selected features and instances. In IBMCCA, two different populations represent different aspects of the classification problem, such as feature selection, instance selection. These populations collaborate and exchange information through a co-operative co-evolutionary framework to find optimal individuals that have both convergence and diversity factors. These individuals are then utilized to generate a collection of subsets of data that are used to generate classifiers in ensemble learning algorithms. Combined with hybrid data resampling methods, IBMCCA has shown a good ability to handle problems related to imbalanced data. Experimental results on 21 datasets and comparisons with many other algorithms have clearly illustrated the efficacy of the proposed algorithm for enhancing the classification performance on imbalanced data.