Research On P2P Personal Credit Risk Based On Machine Learning

With the wide application of Internet finance,P2 P lending has become a new type of financing mode,which greatly improves the utilization efficiency and solution of social funds,solves the problem of personal shortage of funds and forms a truly inclusive finance.Faced with relatively high returns,the high risk of P2 P lending has been criticized by people all the time.How to predict the default risk of credit in advance and identify the key risk features performs a key role to ensure the safety of personal credit behavior,fully mobilize the vitality of the private economy and promote the healthy development of personal credit.With the rapid development of artificial intelligence technology,it is an important and feasible way to effectively use the key technology of artificial intelligence to assist the financial system in decision-making and risk avoidance.This paper focuses on the personal credit risk assessment,especially the personal credit risk assessment of P2 P credit platform.It selects the credit loan data from 2007-2017 of Lending Club,an American online loan platform,and selects 23 field dimension information of 250 thousand records for data processing to predict and classify the borrower's credit risk level indicators.Firstly,the paper uses the classical statistical methods,such as data distribution analysis,correlation coefficient analysis,principal component analysis and so on,to understand the basic characteristics of the data of each field.On this basis,the application of the systematic machine learning method is performed,mainly using the k-nearest neighbor method,decision tree method,neural network method,logistic regression method,support vector machine(SVM)method to make a comprehensive classification prediction of personal credit risk.The experimental results of logistic regression and SVM were 97% and 98% respectively,which achieved a high accuracy of prediction and discrimination.Through the comparison of different methods,it can be found that the effect of regression-type methods are better than that of other machine learning method in this kind of data set,which indicates that there are key characteristic fields such as "borrower's monthly debt","loan term" and "total credit limit" that affect the credit risk rating prediction in P2 P personal credit risk assessment,and at the same time,the integrated key index has a good measurement effect for the credit risk evaluation.In this paper,through the application of statistical and machine learning methods in the actual data analysis,the research results can fully reflect some key characteristics and laws of P2 P credit platform personal risk assessment.The deep integration of artificial intelligence and Internet finance can effectively predict the borrower's risktaking ability and the possibility of default,which is of great significance to ensure the smooth work of P2 P credit and the orderly operation of financial order.