Early Warning of Credit Risk of Internet Financial Enterprises Based on CNN-LSTM Model

Abstract

The importance of enterprise credit risk management is increasingly prominent. Under the background of financial technology and big data, this paper studies a model combining Convolutional Neural Network (CNN) and Long Short-Term Memory Network (LSTM) by comprehensively analyzing multi-dimensional data such as corporate financial statements, credit scores and transaction records, so as to capture and learn the complex characteristics of credit risk of Internet financial enterprises. Firstly, this paper collects the relevant data of Internet finance enterprises from multiple data sources, and carries out standardization processing and missing value filling. Then, CNN-LSTM model is constructed based on these data, and model training and hyperparametric optimization are carried out by adjusting convolution layer and LSTM layer. In addition, this study also designed a comparative experiment to evaluate the performance of CNN-LSTM model and CNN and LSTM models alone in prediction time, prediction accuracy and interpretability. The results show that CNN-LSTM model has significant advantages in predicting the credit risk of Internet finance enterprises. The model also shows a faster response speed, and the maximum warning time is only 701ms, which is much lower than the LSTM and CNN models. The highest accuracy of the model is 94.1%, which is significantly higher than that of LSTM and CNN models. In addition, the model also has high confidence in interpretability, which provides a solid basis for financial decision-making.