A new hybrid method for credit scoring based on clustering and support vector machine (ClsSVM)

The main goal in the credit scoring process is forecasting every customer´s adequacy in accomplishment of their obligations precisely as much as possible. Although this technique is identical with regular binary classification tasks but it has a few crucial differences. Whereas, based on financial credit rules, a customer is considered based on a degree of goodness or badness, one cannot allocate them to one of two distinct classes. Although, in order to solve this problem, researchers have tried to use classification methods that enable producing a posterior probability of default instead of pure classification results, all of them have drawbacks, which cause many serious problems. In addition, the performance of the final model isn´t so high and the error rate is remarkable. In this paper a new hybrid method to address these problems is proposed, which can efficiently increase credit scoring accuracy with no previous assumption. Comparison of the obtained results on proposed hybrid method with Logit model in a real world dataset indicates a significant performance improvement.