Credit Risk Prediction Network Based on Semantic Feature Transformer and CNN

Credit risk is one of the main risks faced by financial institutions. Establishing credit risk prediction model is an important means to help financial institutions avoid risks and reduce losses. To fully extract the feature information in the credit data and improve the parallel computing capability of the model while reducing the network size, a credit risk prediction network based on semantic feature Transformer and CNN is proposed in this paper. Firstly, a Semantic Feature Transformer (SFTransformer) module is constructed by introducing semantic features into Transformer to extract credit risk features. Secondly, because one-dimensional convolutional neural network (1DCNN) has good feature extraction ability and parallel computing ability, the 1DCNN module is combined with the SFTransformer module to build a parallel network CNN-SFTransformer. Finally, the feature data of the borrower is fed into the CNN-SFTransformer model to calculate whether the borrower has default risk. In this paper, two real credit datasets from Australia and German in the UCI database are used for experiments. AUC and KS, two commonly used evaluation indicators, are used to evaluate the predictive performance of the model. Experimental results show that the CNN-SFTransformer model proposed in this paper has good performance in the task of credit risk prediction, with high accuracy and reliability.