The class imbalance problem (CIP) occurs when one class has more instances than other, causing classifiers to lean toward the majority and overlook minority instances. To handle this challenge, this study proposes Cluster Bagging (CLUB) an ensemble-based resampling approach. CLUB applies Fuzzy C-Means (FCM) clustering to divide the majority class into smaller, meaningful clusters and combines each cluster with minority instances to create balanced subsets. Each subset is then trained using the Random Forest algorithm, and final predictions are aggregated through a voting mechanism. Simulations conducted on 73 datasets from the KEEL and UCI repositories reveal that CLUB delivers improved results than 20 leading methods, achieving above 90% G-mean and F1-scores when the imbalance ratio (IR) is below 10.
Kumar et al. (Fri,) studied this question.