A Relabeling Approach to Handling the Class Imbalance Problem for Logistic Regression

Logistic regression is a standard procedure for real-world classification problems. The challenge of class imbalance arises in two-class problems when the minority observed much less than majority class. This characteristic endemic many domains. Work by Owen has shown that cluster structure among may be specific problem highly imbalanced logistic regression. In this article, we propose novel relabeling approach to handle using regression, which essentially assigns new labels observations. An expectation–maximization algorithm formalized serve as tool efficiently computing relabeling. Modeling on such relabeled data can lead improved predictive performance. We demonstrate effectiveness with detailed experiments real datasets. Supplemental materials article are available online.