bank client credit scoring, along with loan parameters optimization using the simulation-optimization model

The present study aims to assess the new method presented for credit scoring and bank loan parameters optimization by simulation-optimization approach. the proposed method contains stages including data preparation, credit scoring, and simulation optimization. during the first stage, the data related to bank loans and nancial statements of companies are collected and the required features are cal- culated. the critical features are selected by the minimum redundancy maximum relevance (mrmr) algorithm. then, the classification methods including logis- tic regression (lr), k-nearest neighbor (knn), artificial neural network (ann), adaptive boosting (adaboost), and random forest (rf) are utilized to solve the credit scoring problem. the performance of these models is evaluated by criteria such as accuracy, f1-score, and area under curve (auc), and the best model is selected for the next stage. during the simulation-optimization, the optimal fea- tures of the loan granted to clients are considered to minimize the default rate of the loan. to this aim, the loan size, interest rate, and repayment period are regarded as variables of the optimization problem. the optimization problem is solved by the memetic algorithm (ma) in four cases. a pre-trained credit scoring model is applied in the ma to estimate the probability of client default. a case study was conducted on the data related to 1000 legal clients of a commercial bank in iran. eleven features were selected to be employed in the credit scoring among the 30 defined. the rf method performed best among the credit scoring models. the simulation-optimization approach reduced the default rate from 38% to 20% by decreasing the loan size and interest rate, as well as increasing its age. the results indicated the efficiency of the proposed method in reducing the credit risk of banks.