Credit risk Evaluation by hybrid data mining technique

Most studies have concentrated on building an accurate credit scoring model to decide whether or not to grant credit to new applicants and the efforts to build more accurate credit scoring model seems to be not significant. In this paper, we proposed a hybrid data mining technique which contains two processing stages. In the clustering stage, the samples of the accepted and new applicants are grouped into homogeneous clusters, the isolated samples are deleted and inconsistent samples are relabeled. In the classification stage, support vector machines used samples with new labels to build the scoring model. The difference from the other credit scoring model is that the samples were classified into three or four classes, rather than two the good and the bad credit classes. Experimental results based on the credit data set provided by a local bank in China showed that by choosing a proper cut-off point, super classification accuracy of the good and the bad credit is obtained. Risk management strategies are developed according to the characteristic of each class.