Integrating MTS with bagging strategy for class imbalance problems

Abstract

Class imbalance is a common problem in classification tasks. The learning schemes of most classification algorithms tend to optimize the overall accuracy, and thus, identification of important but rarely occurring examples is ignored. The Mahalanobis–Taguchi system (MTS) has been shown to be robust in addressing class imbalance problems owing to its inherent properties of classification model construction. The bagging learning approach often has been applied as a superior strategy to reduce the learning bias of classification algorithms. In this study, we propose MTSbag, which integrates the MTS and the bagging-based ensemble learning approaches to enhance the ability of conventional MTS in handling imbalanced data. We perform numerical experiments involving multiple datasets with various class imbalance levels to demonstrate the effectiveness of MTSbag, especially for datasets with high imbalance levels. Finally, as a healthcare application, an early warning system for in-hospital cardiac arrest, was successfully implemented by leveraging the minority class identification ability of MTSbag.