Predicting the Credit Risk of Loans Using Data Mining Tools

One of the most common causes or credit phenomenon that is taken into account for credit risk is the customer’s noncompliance with the commitments. Thus, by predicting the behavior of loan applicants, the growth rate of debts can be decreased. Hence, this study is conducted on corporate applicants for loans in one of the public banks in Iran. In this paper, the main elements comprising the customers’ behavior are selected with the help of categorized sample collection of 521 random samples from all corporate applicants. the process of data preparation, then, is accomplished by summarization, integration, and interpolation of some lost data. In the next step, 85 key performance indicators are selected for modeling. In order to measure the importance degree of the affecting elements on the customers’ behavior, the decision tree، neural net algorithms and Support Vector Machine were applied, the decision tree algorithm with 14 percent average absolute error, having the highest degree, was recognized as the top algorithm capable of assessing the probability of defaults. Finally, based on the available data and according to the results of the CHAID decision tree, contract maturity, amount of interest, number of installments, operating profit to asset, type of contract, average debt  in 3 months ago and loan amount are the most important indicators affecting the customers’ behavior. Taking these indicators into account, before granting the loans, could have a significant role in the prediction of the customers’ behavior and the related decision making.