Nowadays, lending is one of the activities that contributes significantly to the income and profit of banks and fintech companies. Therefore, credit rating/scoring and risk assessment tools play very important roles to minimize credit risk, defined by the client’s inability to repay the loan which the bank has granted. In this thesis, we will propose a novel method which utilizes residual neural networks for credit scoring. The proposed approach converts tabular data sets into black/white images and thus allows the application of a residual neural network with 50 layers (ResNet50) in credit scoring. Each pixel of the image corresponds to a feature bin of the tabular data set. The predictions from the ResNet50 are interpreted using the SHAP method. We did the experiments for the proposed method and logistic regression for two publicly available data sets of different sizes. Based on the results, our proposed approach shows superiority compared to logistic regression, especially, if the sample size is large.     «

Nowadays, lending is one of the activities that contributes significantly to the income and profit of banks and fintech companies. Therefore, credit rating/scoring and risk assessment tools play very important roles to minimize credit risk, defined by the client’s inability to repay the loan which the bank has granted. In this thesis, we will propose a novel method which utilizes residual neural networks for credit scoring. The proposed approach converts tabular data sets into black/white images...     »