Research On The P2P Lending Platform Credit Risk

The credit risk of P2 P online lending platform in this paper refers to the credit risk events that take P2 P online lending platform as the main body,such as delayed payment,difficulty in withdrawing cash,intervention of economic investigation,platform shutdown,etc.To regulate the development of the Internet financial in our country,the protection of the interests of investors,optimize the market supervision system,in this paper,the research of P2 P network platform loan credit risk,all kinds of platform according to the different credit risks by the hierarchy,screening and analysis of the characteristics of all kinds of platform and credit risk characteristics,to guide investors to make decisions,and provide early warning and advice to regulators.The perspective of data mining based on which this paper is based specifically refers to the two most classical algorithms in data mining technology at present--unsupervised learning and supervised learning.Unsupervised learning mainly uses k-means clustering and hierarchical clustering,and supervised learning includes support vector machine and decision tree.Compared with traditional credit rating methods(in this paper,it specifically refers to expert rating method,principal component analysis method and analytic hierarchy process),data mining method has three advantages.Second,the machine learning algorithm can directly explore the internal rules and associations of data points based on the internal connections of data.Without relying on the previous research experience,it is possible to obtain extraordinary results.It is also more suitable for the rapidly developing and significantly dynamic P2 P online loan rating industry in China.Thirdly,the machine learning algorithm is simple to call and fast to model,which makes it convenient to compare the accuracy of different types of lifting algorithms.Moreover,with the expansion of training data,the prediction accuracy will continue to improve.This paper use the unsupervised learning methods to collect 268 platform,15 indexes according to cluster analysis,get the biggest difference between each type and each type of each platform is similar to the greatest degree of five types,and analyzesthe main characteristics of each type,for different level of credit risk classification of the largest contribution to the key indicators.The second step using supervised learning method,first using the SVM results validate the unsupervised learning methods,and then through the has labeled data,including normal,economic intervention,delay to deal with the three types of platform,training for supervised learning model,and compared the SVM,ID3 decision tree,CART decision tree C4.5 decision tree and the applicability of the model in this area,the use of these models can be no labels for other data to identify and differentiate,achieve a certain degree of risk prediction function.