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

Abstract Real-world classification domains, such as medicine, health and safety, finance, often exhibit imbalanced class priors have asynchronous misclassification costs. In cases, the model must achieve a high recall without significantly impacting precision. Resampling training data is standard approach to improving performance on binary data. However, state-of-the-art methods ignore local joint distribution of or correct it post-processing step. This can causes sub-optimal shifts in distribution, particularly when target complex. this paper, we propose Radial-Based Combined Cleaning (RB-CCR). RB-CCR utilizes concept potential refine energy-based resampling CCR. particular, exploits accurately locate sub-regions data-space for synthetic oversampling. The category sub-region oversampling be specified an input parameter meet domain-specific needs automatically selected via cross-validation. Our $$5\times 2$$ 5 × 2 cross-validated results 57 benchmark datasets with 9 classifiers show that achieves better precision-recall trade-off than CCR generally out-performs terms AUC G-mean.