Research On Credit Risk Assessment Of Internet Lending Based On GA-BP Model

The P2 P network lending industry has also entered a high-speed growth period along with the rapid development of the Internet in recent years,which gradually fills the gap in the traditional lending industry and has become one of the important channels for small,medium-sized enterprises and personal lending.Improving the risk identification capabilities of P2 P online lending platforms and enhancing the ability to judge borrowers' credit risk become the primary task of my country's P2 P online lending platforms.This thesis explores machine learning algorithm to predict the borrower's default behavior through borrowers' historical loan data,and uses a reasonable way to give a mark about the user's credit.These details are as follows:First,this thesis explains the development and current situation of the P2 P online lending industry,and expounds the main research content of this article based on the GA-BP model of online lending credit risk assessment research.Secondly,this thesis researches the relevant theory of BP neural network and genetic algorithm,and constructs BP neural network model(GA-BP model)based on genetic algorithm optimization.Thirdly,choosing the loan data of the Lending Club from January 2017 to June 2018,this thesis performs the user's default profile analysis based on the user's housing nature,working years,loan purpose,annual income and other variables,analyzes the the link between default risk and user characteristics.Then,this thesis comes up with the original feature derivation method to solve the problem that the features of the original data set cannot utterly draw the borrower's information.At the same time,it explores the related theory of random forest to mining the feature importance screening,and finally selects the feature set for model training according to Pearson correlation coefficient.The results of the BP neural network test gives the conclusion that the model trained the data set containing derivative features can effectively improve the prediction accuracy.Finally,the GA-BP model predicts using the data set obtained above after training.The results show that the prediction effect of the GA-BP model is better than the traditional BP neural network.Then,this thesis proposes the method of using the output layer sigmoid function value of the GA-BP model as the borrower's default probability to make a credit score and establish a score card.It also gives corresponding loan suggestions in different credit scoring ranges,which has a certain reference value for P2 P network lending platforms and makes a certain positive impact on the development of my country's P2 P network lending industry.