Classification Methods of Credit Rating - A Comparative Analysis on SVM, MDA and RST

The execution and the result of bank credit rating are closely linked with the bank´s investment and loan policies which form the initial risk measurement. It is an important and a shouldn´t ignored issue for bankers to set up a scientific, objective and accurate credit rating model in the field of customer relationship management. In this study, two classification methods, multiple discriminate analysis (MDA), CANDISC, and support vector machine (SVM) are applied to conduct a comparative empirical analysis using real world commercial loan data set. The result comes out that SVM model has reliable high classification accuracy under feature selection and therefore is suitable for bank credit rating. This study suggests the decision-making personnel to establish a decision-making support system to assist their judgment by using the classification model.