Development and Explanation of Bank Customers' Credit System Based on Hybrid Learning Models: A Case Study of Bank Mellat

The possibility to renewal of loan contracts in Iran may lead to the identification of fictitious profits by banks and ultimately lead to a banking crisis and disruption of the country's monetary system, so to prevent banks from reaching the stage of this, Measuring customers' credit risk is essential. The aim of this study is to increase the accuracy of customer accreditation using the structure of hybrid models and has been done as a case study on Mellat Bank. In this regard, 14 learning models were compared with each other and their ability to validate customers was determined. Learning models show that based on both accuracy criteria (success rate) and measurement F (harmonic mean between accuracy and recall), the combined learning model (KNN-NN-SVMPSO) - (DL) - (DBSCAN) with an accuracy rate of 99.90 is the highest It has more performance than other basic and hybrid models. Also, using the principal component analysis (PCA), Gini index, interest rate ratio (IGR) and interest rate (IG) methods to calculate the weight of features and their average rank, it was shown that features such as collateral, The type of collateral and the amount of facilities have been the most important features in distinguishing good from bad customers, respectively.