| With the rapid development of the Internet,the traditional financial industry has been impacted.Such an environment has promoted the birth of P2 P online lending platforms.Users do not need to borrow through intermediaries,but directly through the Internet to achieve person-to-person lending behavior.P2 P platform is relatively wide in object orientation,with low requirements and easy to operate.More and more people choose to borrow money on P2 P platform,but the consequent problems are gradually emerging.Because of information asymmetry,lending to the P2 P network platform is difficult to fully grasp the information of the user,credit assessment of the difficulties of borrowing threshold is low,the customer information is not comprehensive,loan amount is less,the phenomenon such as default more increased the difficulty of credit assessment,because the client every day a lot of transaction data,traditional way of credit evaluation is not applicable,establish a more efficient,more accurate and more objective,more low-cost credit evaluation system become a difficult problem.This paper takes the user data of Lending Club in the first half of 2019 as the research sample and the default customer identification algorithm as the research object.The research content mainly is divided into five parts: First,explain the research background and significance of the paper,summarize the research status of domestic and foreign scholars on the impact factors of P2 P lending platform default and default prediction model,and introduce the research framework of paper.Second,it introduces the research on P2 P online lending,summarizes the stages that P2 P online lending platforms have gone through in China and the operation modes of P2 P online lending in China,and summarizes the main risks of P2 P online lending.The classification problem and integration learning strategies are briefly introduced,and the basic principles of XGBoost algorithm,random forest algorithm,voting classification algorithm and algorithm evaluation index are introduced in detail.Thirdly,statistical analysis and data preprocessing are carried out.Fourth,XGBoost,random forest and voting classification algorithm are used for empirical analysis,and comparison and scoring with other algorithms.Fifth,summarize the research results and put forward relevant Suggestions.The results show that: in terms of the factors affecting the default,characteristic variables such as occupation,annual income,working years,loan amount,loan purpose and housing ownership status all have a certain impact on the default risk of customers."The number of circular trade opened in the past 12 months","loan interest rate","mortgage account number","current active transaction number" and other economic behaviors have a relatively high contribution degree,which play a very important role in the default customer identification model.In terms of default recognition algorithm,the random forest algorithm(0.92955)scored the highest in F1,the k-nearest neighbor algorithm(0.96624)scored the highest in recall,the XGBoost algorithm(0.99157)scored the highest in precision,and the random forest algorithm(0.97820)scored the highest in AUC.The highest overall score was achieved by the random forest algorithm,followed by the XGBoost algorithm and the voting classification algorithm.The performance of the integrated learner is much better than that of the individual learner,but the overall operation time is long,the parameter adjustment is difficult and the cost is high.According to the research results,the following Suggestions are put forward: 1.Improve the risk control of online loan platform,and scientifically assess the default risk of customers;2.Improve the customer credit audit mechanism,accelerate the pace of the whole society;3.Strengthen the management of P2 P industry and improve the legal standards of Internet finance;4.Closely follow the domestic Internet financial policies and define the development direction. |
PDF Full Text Request