Application Research On Loan Risk Measurement Of P2P Network Based On L1 Regularized Logistic Regression Model

P2P network loan is a kind of Peer-to-Peer borrowing behavior.It acts as an intermediary for the entire lending link,providing borrowing services for borrowers and investors,and charging the corresponding fees as a profit model.The loan method innovatively integrates the capital market of modern private lending business,Internet artificial intelligence information technology,financial service project and electronic information commerce platform,and borrows the small capital that is gathered together in the form of similar bonds through the Internet platform.There are people in need or small and medium-sized enterprises.By the end of 2018,there were more than 6,000 platforms similar to online lending functions.These platforms are different from traditional formal financial services,and they are easy to obtain large amounts of capital and profits.Due to factors such as personal credit problems and timely implementation of policies,the borrower has not repaid on time,and the platform intermediary has “running the road”,making the field a high-risk industry.In 2018,there were 4,672 abnormal platforms in China,accounting for 77.4% of the total number of platforms.This paper takes the cause of the credit risk of online loans as an entry point,deeply analyzes the causes of its risks,and combines empirical analysis to make policy recommendations for China.Due to the small scale of domestic P2 P online loans and the low integrity of data,this paper is aimed at the current largest P2 P network loan platform in the world,Lending Club.In the study,a total of 1,340,797 samples were collected from 2016 to 2018 for three years,of which 43041 were valid samples.The article is divided into four modules to elaborate.Firstly,it introduces the related theories of credit and risk,as well as the problems existing in China's P2 P online lending industry and the government's rectification measures.Secondly,it analyzes the machine learning thought and its core technology.According to the collected data samples,the advantages and disadvantages of different machine learning models are discussed and compared.Finally,the Logistic regression model is selected to study the P2 P network loan risk measurement.Thirdly,in order to find the optimal model,the L1 and L2 regular terms are also introduced to optimize the model.Through the study of machine learning methods,it was found that the risk assessment effectiveness of L1 regularized Logistic regression(AUC=0.838)was significantly better than traditional Logistic regression(AUC=0.740)and L2 regularized Logistic regression(AUC=0.791).The actual results are in line with the theoretical situation,indicating that the Logistic regression model is feasible and reliable for P2 P network loan risk measurement research.Finally,it concludes that the risk measurement direction is the necessary condition for the steady development of the P2 P online loan industry.At the same time,it proposes policy construction for the future P2 P network loan business,which has very important reference value for the improvement of P2 P network loan service system in China.