A bank credit risk model for customer classification based on the evolutionary crow algorithm and convolutional neural network

Creditworthiness is one of the most vital and important tasks in the modern banking industry, which ensures the non-performing and sustainability of credit financial institutions with an accurate prediction of the credit status of loan applicants.The main goal of this research is to improve credit assessment for small companies, which is an innovative measure in this field.In fact, to achieve this goal, comprehensive datasets including a variety of financial information, micro-company information, public credit information, and third-party personal access to information are required so that unbalanced sample processing techniques can be used to achieve fairer samples. In this regard, a crow algorithm is used to solve the feature selection problem and in the second step, the selected features are provided as input to a convolutional neural network for final classification. The proposed model was implemented and evaluated on a real-time validated dataset from the Kaggle platform, which included real banking information. K-Fold cross-validation method was used to accurately measure the model performance. The experimental results show that this hybrid model has achieved significantly higher accuracy compared to traditional classification methods. Finally, this research showed that the proposed model has high potential for use as a decision support system in banks and financial institutions for credit assessment.