FARCI+: Enhancing Fuzzy Rule-Based Classification for Imbalanced Problems via Choquet Integral Generalizations and Support Tuning

Imbalanced classification problems pose a significant challenge in machine learning, especially when the minority class contains critical information. In this context, Fuzzy Rule-Based Classification Systems (FRBCSs) have been widely used due to their interpretability and flexibility, but typically rely on sampling techniques to address class imbalance. FARCI, a fuzzy association rule-based classifier, is specifically designed for imbalanced datasets by introducing tailored modifications to the FARC-HD fuzzy classifier, improving performance without relying on data preprocessing. However, FARCI uses fixed membership function supports, which may limit its effectiveness in small disjuncts, and the application of the additive combination may introduce bias during inference. In this paper, we propose FARCI+, an enhanced version of FARCI that addresses these limitations through two key contributions: (1) it tunes the support of the membership functions using the 3-tuples fuzzy linguistic model to better represent minority class subregions, and (2) it incorporates generalizations of the Choquet integral in the inference stage to achieve more robust aggregation in imbalanced contexts. The effectiveness of FARCI+ is assessed through extensive experimentation on 66 binary imbalanced datasets from the KEEL repository. The results demonstrate that the combination of both components in FARCI+ significantly improves classification performance, particularly for datasets with high imbalance ratios, without affecting performance in moderately imbalanced datasets.