Rough Sets in Imbalanced Data Problem: Improving Re-sampling Process

摘要

Imbalanced data problem is still one of the most interesting and important research subjects. The latest experiments and detailed analysis revealed that not only the underrepresented classes are the main cause of performance loss in machine learning process, but also the inherent complex characteristics of data. The list of discovered significant difficulty factors consists of the phenomena like class overlapping, decomposition of the minority class, presence of noise and outliers. Although there are numerous solutions proposed, it is still unclear how to deal with all of these issues together and correctly evaluate the class distribution to select a proper treatment (especially considering the real-world applications where levels of uncertainty are eminently high). Since applying rough sets theory to the imbalanced data learning problem could be a promising research direction, the improved re-sampling approach combining selective preprocessing and editing techniques is introduced in this paper. The novel technique allows both qualitative and quantitative data handling.