ARF: A hybrid model for credit scoring in complex systems

Numerous quantitative models have been developed for credit scoring of bank customers, by which banks assess the risk of their credit customers. These models mainly determine customer rating using previous trends. Due to the changing economic conditions and various social and political developments and its impact on the economy and economic enterprises, quantitative models alone do not have the appropriate accuracy and there is a need for a comprehensive model taking into account quantitative and qualitative conditions and expert opinions. In this article, 5C criteria have been used to score the customer, which are: character, capacity, capital, collateral and conditions. The customer condition criterion is highly affected by economic shocks. The status of the corresponding listed companies has been used to identify and determine the impact of the customer's economic shocks. A hybrid model is presented that detects and predicts the shocks of different stock market segments based on adaptive neum-fuzzy inference systems (ANFIS) and recurrent neural network (RNN) using historical data and indicators. Then the results along with other customer criteria are entered into a fuzzy rule base (FRB) which finalizes the customer score. Based on the obtained results and pattern, it is possible to repay the loan on time for the customers who are expected so, and for suspicious customers, the loan will be prevented or more closely monitored.