Imbalanced Data Classification via a Cost-Sensitive Majority Weighted Minority Oversampling Approach

摘要

To address the imbalance classification tasks, a method for Cost-sensitive Majority Weighted Minority Oversampling (CS-MWMOS) technique was proposed. The betweenclass and inner-class imbalance problem of the sample distribution are,synthetically considered. Imbalance ratio was introduced to alleviate between-class imbalance issue. The inner-class imbalance problem is also reduced by allocating different weights for minority sample. Extensive experiments on 20 UCI imbalance datasets showed that the proposed method can effectively address the class imbalance problem in terms of the geometric mean and average accuracy assessment metrics.