Better classifiers for credit scoring: a comparison study between self organizing maps (SOM) and support vector machine (SVM)

Credit scoring has become an increasingly important area for financial institutions. Self Organizing Maps (SOM) and Support Vector Machine(SVM) are two techniques of data mining which are being used in different applications of businesses. In this paper, descriptive variables in literatures and criteria are being used, which affect the credit of customers in the Iranian financial institutions. We begin with evaluating these variables using Multi Criteria Decision Making (MCDM) approach and take into account the psychological and social viewpoints of the experts. Next both SVM and SOM methods are applied to the credit database and the results are compared. To compare these two methods we use coincidence matrix and the Type I and Type II errors. We show that they are competitive and most significant in determining the risk of default on bank customers. In this paper 2 standard formulated methods and one Innovative algorithm based on SVM and SOM were applied to classify the customers. The results reveal that proposed model performs significantly better than standard SOM and SVM. Additionally the proposed model solves one of the most important challenges in our research which is the ability to detect bad customers.