Co2RBFN-CS: First approach introducing cost-sensitivity in the cooperative-competitive RBFN design

Abstract

The interest in dealing with imbalanced datasets has grown in the last years, since they represent many real world scenarios. Different methods that address imbalance problems can be classified into three categories: data sampling, algorithmic modification and cost-sensitive learning. The fundamentals of the last methodology is to assign higher costs to wrong classification classes with the aim of reducing higher cost errors. In this paper, CO2RBFN-CS, a cooperative-competitive Radial Basis Function Network algorithm that implements cost-sensitivity is presented. Specifically, a cost parameter is introduced in the training stage of the algorithm. This parameter modifies the learning rate of the weights taking into account the right (or wrong) classification of the instance, the type of class (majority or minority) and the imbalance ratio of the data set.