Investigating the Variables Affecting Banks’ Legal Customers Credit Risk, Using Support Vectors Machine and Decision Tree

The increase of non-current debts to lending facilities ratio as an indicator of banks' credit risk can endanger the health of the banking sector, financial system and the real economy. Hence, in this paper, analyzing credit risk through the actual balance of non- performing debts by focusing on a broad set of variables including financial, non-financial, contractual, audit and economic variables in a sample of 677 legal customer facility files of a State Bank for the years 2006- 2017 has been accomplished. Based on the results, the LASSO Algorithm with better performance has identified 10 key financial, economic and audit variables affecting the credit risk. However, training these features by support vector machine and decision tree model, which represent the best results in the Lasso algorithm with the decision tree application, confirms the small significance factor for the audit variables. Therefore, using LASSO algorithm with decision tree and focusing on financial and economic variables can be sufficient for credit risk analysis.