Improving Feature Selection for Credit Scoring Classification Using a Novel Hybrid Algorithm

Abstract

Credit scoring has become a very important issue and has been studied extensively in financial companies and banks. However, in credit scoring datasets, a large amount of irrelevant and redundant features are involved, which decrease the classification accuracy. Consequently, the effective feature selection approaches are become a necessary approach. In this paper, a hybrid feature selection algorithm that combines the genetic algorithm (GA) and the backpropagation neural network (BPNN) classifier is proposed. With the hybridization, the GA works to select the subsets of characteristics through the process of feature selection (FS) and then the BPNN evaluates the selected subsets by a fitness function. According to three bench mark credit scoring datasets, the experiment results demonstrate that the proposed hybrid approach has a superior performance in terms of evaluation criteria compared with other competitor approaches.