Empirical Research On The Default Risk Prediction Of Peer-to-peer Lending Platform Based On Machine Learning

As a new financial service model,P2 P network lending is more efficient and convenient than traditional banks in solving micro-enterprise financing and personal loans.It eliminates a series of cumbersome procedures of off-line private loans and bank loans and provides online transactions directly,which makes users greatly enjoy the convenience brought by the virtual world.This new network lending model has been widely welcomed.However,in recent years,the risk of online lending has accumulated and a series of risk events have erupted,resulting in the bankruptcy of a large number of platforms,which not only seriously damages the legitimate rights and interests of investors,but also greatly endangers the security of the financial industry and social stability.For this reason,the online lending platform department has also issued corresponding policies to guide the regular and healthy development of this industry.Besides,for the platform itself,risk management and early warning should also be done well.Because there are many default risks in the P2 P network lending platform in China,it is urgent for the current P2 P network lending platforms to accurately predict default risks and take appropriate measures to deal with them.This paper mainly uses several machine learning methods to construct models to study the default rate of P2 P users.Compared with each model,the most suitable risk identification model is found out for the default rate of P2 P users,and helps the P2 P industry to establish a more suitable risk identification system.The research on risk prediction of users in P2 P can help the platform identify the borrowers who may most likely default in advance,and protect the interests of the platform itself and its users.On the basis of previous research,this paper selects original lending loan transaction data of users from 2016 to the first quarter of 2017,which is publicly provided by Lending Club,to carry on the empirical study on the risks of loan default by respectively using logistic regression model,the bat optimization algorithm of feedforward(BAPA)neural network and least squares support vector machine(LSSVM)to analyze the experimental data.The focus of this paper is to construct three models,and then apply them to the default risk prediction experiment based on these three methods.Lastly,the applicability of the three methods in predicting the default risk of P2 P network lending platform is evaluated by comparing the results calculated by the three models.The experimental results of each model show that the least squares support vector machine(LSSVM)has better prediction effect and is most suitable for online lending platforms to use as default risk identification model.In this paper,logistic regression model is completed by binary logistic regression analysis in SPSS statistical software,and BP network and LSSVM are implemented by compiling corresponding model code in MATLAB.