Study on Credit Default Risk Prediction Model Based on BP-RF Neural Network

Abstract

In the Internet financial industry, it is of great significance to the user's credit default risk management, but the traditional machine learning model has low prediction accuracy. This paper proposes a two-stage credit default identification model based on BP-RF (Back Propagation Neural Network with Random Forest). Firstly, the feature of transaction data is extracted automatically by constructing the neural network, and then classified and predicted by the random forest algorithm. The model is verified with millions of data sets provided by the lending club. The results show that the model has better performance than the traditional model, with an accuracy of 95.1%, AUC index of 89.0% and ACC index of 93.1%, which proves that the model is more suitable for the field of credit default prediction.