USING SUPPORT VECTOR MACHINES FOR THE COMMERCIAL BANK CREDIT RISK ASSESSMENT

In this article, we introduce a different method called the overall risk measure in the bank which is different from the "enterprise-bank" internal measure from the micro system. This method measures the credit risks of an enterprise in a macro perspective. Specially, we take the ratio of non-performing loan (NPL) as the indicator to measure credit risks and use the support vector machines (hereinafter referred as: SVM) to predict it, as the support vector machines have advantages in processing high-dimension samples. Furthermore, we choose some indicators relevant with ratio of non-performing loan and use the principle component analysis and the recursive feature elimination to exclude indicators. And then support vector regression (hereinafter referred as: SVR) train samples to output ratio of non-performing loan. When comparing all kinds of ways to choosing indicators, we find out a regression model combined by deleting insignificant indicators and recursive feature elimination has the better prediction accuracy than the BP neural network. Finally, the empirical results evidence that SVR has a good prediction.