Effective Banking Supervision in Granting Micro-loan with Emphasis on the Credit Rating of Real Customers

The present research with emphasis on the use of the combined model of logistic regression, artificial neural networks, and symbolic regression, the current research aims to identify the factors that affect the credit risk of real customers in selected branches of Sepah Shiraz, with the aim of effective monitoring of micro-loan. By examining a sample of 351 real customers, 17 variables included; loan amount, repayment term, interest rate, income, age, number of bounced checks, debt history, account lifetime, type of collateral, education, gender, spouse's employment, marital status, property status, job, type of loan, obligatory or non-obligatory status for the classification of customers into good and bad accounts were categorized and extracted. Using forward selection technique of parent, 5 variables affecting credit risk were selected and used to train a neural network with three neurons in the hidden layer. The optimal cutting point was selected based on the characteristic curve of the system performance. The results of the output of the artificial neural network on the test data showed that the accuracy of the combined model of logistic regression-artificial neural networks in the classification of good customers is 0.89 and in the classification of bad customers is 0.83. The results of logistic regression, and the combined logistic regression-symbolic regression model is better.