Research On P2P Network Loan Customer Default Prediction

With the rapid development of Internet finance and the increase of residents 'income,residents' financial management concepts and spending habits have also changed.The concept of Internet financial management has become increasingly popular with the public.The arrival of big data era makes the information technology gradually improved,P2 P online loan platform has become the main way of mass financial management.P2 P network loan platform not only to develop high-quality financial products,make recommendations for investors,but also to reflect the development of various industries,while P2 P network loan platform can also access the loan customers in many aspects of the relevant data,timely adjustment of the platform's own approval rules And optimize customer-related classification forecasting model to reduce the customer default rate and promote the healthy and stable development of P2 P online loan industry.In this paper,we first make use of P2 P network loan customer data to construct neural networks,classification regression tree and XGBoost model,and evaluate these models.Secondly,based on the constructed model,we analyze the importance variables that affect the customer's default.Second,an example is given to illustrate the method of using the constructed model to predict the default of P2 P network loan customers.Finally,based on the results of the study,this paper provides some suggestions for P2 P online loan platform.The main work of this article is summarized as follows:1,P2 P network loan customer data pretreatment.Through the statistics of the lack of attributes of the ranks,the use of interpolation,median filling;variance stable variable to be removed;outlier elimination;geographic location information and the case of characters such as uniform treatment;2,descriptive statistical analysis of P2 P network loan customer data.Draw a graph of daily loan amount and default status of the loan customers,analyze the business conditions of the P2 P network loan enterprises and the customer default status;draw a map of China by using geographical location information of loan customers to visually analyze areas with high default rates among loan customers.3,P2 P network loan customer data model construction.Firstly,the processed data are integrated and then split into rows to form a training set and test set.The neural network,CART model and XGBoost model are constructed,and the model evaluation index-AUC value is used to compare with the model.It shows that XGBoost algorithm is superior to P2 P Internet loan customers have strong classification prediction performance.4,XGBoost algorithm model.Predict the default situation of P2 P network loan clients and output the variables that have a significant impact on the default of P2 P network loan clients.5,Finally,it summarizes to provide some suggestions for P2 P network loan platform in the classification forecasting method,information construction and sound credit information system.