RB-CCR: Radial-Based Combined Cleaning and Resampling algorithm for imbalanced data classification

Real-world classification domains, such as medicine, health and safety, and\nfinance, often exhibit imbalanced class priors and have asynchronous\nmisclassification costs. In such cases, the classification model must achieve a\nhigh recall without significantly impacting precision. Resampling the training\ndata is the standard approach to improving classification performance on\nimbalanced binary data. However, the state-of-the-art methods ignore the local\njoint distribution of the data or correct it as a post-processing step. This\ncan causes sub-optimal shifts in the training distribution, particularly when\nthe target data distribution is complex. In this paper, we propose Radial-Based\nCombined Cleaning and Resampling (RB-CCR). RB-CCR utilizes the concept of class\npotential to refine the energy-based resampling approach of CCR. In particular,\nRB-CCR exploits the class potential to accurately locate sub-regions of the\ndata-space for synthetic oversampling. The category sub-region for oversampling\ncan be specified as an input parameter to meet domain-specific needs or be\nautomatically selected via cross-validation. Our $5\\times2$ cross-validated\nresults on 57 benchmark binary datasets with 9 classifiers show that RB-CCR\nachieves a better precision-recall trade-off than CCR and generally\nout-performs the state-of-the-art resampling methods in terms of AUC and\nG-mean.\n