Credit Default Prediction Modeling: An Application of Support Vector Machine

Credit default prediction (CDP) modeling is a fundamental and critical issue for financial institutions. However, the previous studies indicate that the classifier’s performances in CDP analysis differ using different performance criterions on different databases under different circumstances. The performance assessment exercise under a set of criteria remains understudied in nature, on the one hand, and the real–scenario is not taken into account in that a single/very limited number of measure only are used, on the other hand. These problems affect the ability to make a consistent conclusion. Therefore, the aim of this study is to address this methodological issue by applying support vector machine (SVM)-based CDP algorithm by means of a set of representative performance criterions, with enclosing some novel performance measures, its performance compare with the results gained by statistical and intelligent approaches using six different types of databases from the credit prediction domains. Experimental results show that SVM model is marginally superior to CART with DA, being more robust than its other counterparts. In consequence, this study recommends that the supremacy of a classifier is linked to the way in which evaluations are measured.