Research On Credit Evaluation Of P2P Platform Based On PSO-BP Model

P2P platform is the Internet finance peer-to-peer lending platform,which originated from a peer-to-peer computer network.The main mode is to accumulate small amounts of funds to lend to those in need of funds through the Internet platform.The first P2P platform in China was born in 2007,and it has been developed to today for nearly 11 years.The number of platforms has exceeded more than 6000,and the loan scale is over about six trillion yuan.However,there are many problems behind the prosperity,the lack of credit audit system and the mode of attracting investors with high interest rates,which leads to the crisis of the platform run.Most of the platforms are faced with bankruptcy,escaping,and cannot be presented,causing significant economic losses of investors,which seriously affected the financial order and social stability of our country.It is particularly important to establish a mechanism to assess and prevent the credit risk of the platform,and to help investors screen a number of good reputable platforms,both from the perspective of economic development and market regulation.The original credit evaluation models are mostly based on qualitative analysis,which are inefficient and susceptible to subjective factors.Later,many scholars put forward quantitative models to assess risks,such as Logistic regression analysis,discriminant analysis and other traditional theories.With the rise of computer technology and artificial intelligence,the theory of machine learning with data driven as the core is concerned by the academic circle.A large number of new modern models have been used in credit evaluation,such as decision tree,neural network,genetic algorithm,random forest and so on,which have achieved good results and are widely used.The purpose of this study is to explore the problem of P2P platform credit evaluation.First,we introduce the advantages and disadvantages of the mainstream Logistic regression,discriminant analysis,decision tree,neural network model and its application,and put forward their own views on traditional statistical methods based on prior knowledge and modern model based on data driven.Then,the basic theory of PSO(particle swarm optimization)is introduced,and a credit risk assessment model based on PSO algorithm and BP neural network model is introduced.Finally,in the empirical analysis part,we collected 220 P2P platforms' data manually from the authoritative third party platform,and processed data by scientific means.The sample data is used to test the prediction accuracy and generalization of the PSO-BP neural network model,and is compared with the mainstream credit evaluation model.The results show that:from the classification accuracy of the model,the classification accuracy of the BP neural network model based on.PSO optimization has not been improved effectively compared with the multi-layer perceptron and decision tree model,but it is all higher than the traditional Logistic model.From the model reliability(prediction accuracy),the traditional Logistic model is obviously better than the multi-layer perceptron and decision tree model,but the reliability of the BP model has been greatly improved after the optimization of the PSO algorithm,and it is higher than the Logistic regression model.Therefore,according to the accuracy and reliability of the comprehensive model,the BP neural network model based on PSO algorithm is the best.