Research On Personal Network Loan Behavior Based On Decision Tree And Random Forest

In recent years,the concept of "Internet Plus" has been vigorously promoted,and the Internet financial industry has developed rapidly,becoming one of the most important innovation projects.From the rise of payment platforms to the online sales of products such as funds,wealth management,and insurance,followed by the launch of personal credit loans and online platform loans,Internet finance has had a huge impact on traditional financial institutions.In addition to the impact of the purchase and loan restriction policies,traditional commercial banks have also launched their own online credit loan products.The environment of China's online lending industry has gradually improved under national supervision,but a large number of defaults by borrowers are still important factors affecting the development of the online lending industry,and the credit assessment methods of most domestic online lending platforms are relatively simple and cannot be effective.To avoid the occurrence of borrowers' defaults,the credit management of borrowers needs to be strengthened.In contrast,the personal credit market in the United States is well developed,and Lending Club,as its representative,is the most successful online lending platform in the world.Therefore,based on the random forest model and the data from the website of Lending Club,this paper constructs a borrower default behavior model to study the influencing factors of borrowers' default behavior in the online Lending platform.In the process of studying the influencing factors of borrower's default behavior,this paper selects the data of borrowers in the first quarter of 2019 from the Lending Club platform for research.Combined with the data produced by the process and significance of the property itself,for the original data of 141 useful attribute has carried on the classification,then using the hierarchical clustering method and principal component analysis(PCA)to conduct feature reduction and feature selection of borrowers' credit attributes with high dimensionality,and use decision tree algorithm and random forest algorithm to construct classification models for the dimensionreduced data.During the comparison and analysis of the models,the superiority of the random forest algorithm in the classification and prediction of unbalanced data was found.and the influencing factors of the borrower's default behavior were summarized,and suggestions were put forward for the review of online lending platform,and a new idea was provided for the classification research of credit data.Studies have shown that the creditworthiness of borrowers in the past is a very important influencing factor for the occurrence of borrower defaults,including the balance of the borrower's account,the maximum balance of the account,the total borrower's loan,the total credit card limit information and other attributes,is an important indicator for judging whether the borrower will commit a breach of contract.The next most important attribute is the natural information of the borrower,including the location of the borrower and the annual income of the borrower.In addition,whether the borrower applies for a joint loan does not help determine whether the borrower will default.Therefore,strengthening the verification of the borrower's credit data at the review stage can effectively avoid the occurrence of defaults.