Radial-Based Undersampling Algorithm for Classification of Breast Cancer Histopathological Images Affected by Data Imbalance

The problem of data imbalance remains one of the most widespread challenges of the contemporary machine learning, affecting most of the real datasets. It is especially ubiquitous in the medical domain, in which some of the classes used during classification can be naturally under-represented. At the same time it can significantly affect the classification performance of most standard algorithms, such as convolutional neural networks. One of the approaches for dealing with data imbalance is artificially reducing the number of objects from the over-represented class, an approach called undersampling. In this paper we present a novel undersampling technique, Radial-Based Undersampling (RBU), and apply it to the classification of breast cancer histopathological images. The results of the conducted experimental analysis not only indicate the usefulness of the RBU algorithm in that task, but also shed light on some of the drawbacks of that method, which could potentially be used in the future to improve upon the proposed algorithm.