Comparative Analysis Of Personal Credit Evaluation Based On Random Forest And Back Propagation Neural Network

With the development of Internet finance,people are no longer satisfied with the traditional banking and other financial institutions for lending and investment,but gradually turn to online lending.At the same time,P2P online lending is an important part of supporting people's demand for lending and investment.In this context,individuals Credit is particularly important,and credit risk has become the biggest risk that cannot be ignored.This paper focuses on personal credit risk assessment,selects the ensemble classifier Random Forest and error back-propagation neural network to build the credit risk assessment model,the random forest algorithm can better tolerate noise,is not easy to produce over fitting,and has high stability,for the personal credit data with high feature dimension and different types,the random forest algorithm can be better performed compared with the traditional single classifier model.Error back-propagation neural network has strong learning ability,and it can provide high classification accuracy for complex credit data,but its disadvantage is poor stability.This paper uses datasets of Lending Club which is free available in the United States.Through a series of data cleaning,transformation,screening and other preprocessing methods,the data has constructed the personal credit risk assessment model based on Random Forest and neural network,and compared with the Logistic Regression model,found that the three models have advantages and disadvantages in credit risk assessment,and then combined the three models into a new model by voting.The result shows that the effect of combined model classification is better than that of three single models.Then we also test the effect of different resampling methods on the model.The results show that the oversampling method has a good effect on this kind of sample imbalance problem.At the same time,there are still many shortcomings in this paper,and there is still room for further improvement in the aspects of feature selection and classification accuracy,which needs further improvement in the future research.