Imbalanced classification by learning hidden data structure

Approaches to solve the imbalanced classification problem usually focus on rebalancing the class sizes, neglecting the effect of the hidden structure within the majority class. The purpose of this article is to first highlight the effect of sub-clusters within the majority class on the detection of the minority instances and then handle the imbalanced classification problem by learning the structure in the data. We propose a decomposition-based approach to a two-class imbalanced classification problem. This approach works by first learning the hidden structure of the majority class using an unsupervised learning algorithm and thus transforming the classification problem into several classification sub-problems. The base classifier is constructed on each sub-problem. The ensemble is tuned to increase its sensitivity toward the minority class. We also provide a metric for selecting the clustering algorithm by comparing estimates of the stability of the decomposition, which appears necessary for good classifier performance. We demonstrate the performance of the proposed approach through various real data sets.