Knowledge Discovery Of Vehicle Credit Data Based On Decision Tree Ensemble Learning

Along with the development of interconnection network technology in recent years,an increasing number of people accept and begin to realize all aspects of life needs through the Internet platform,including learning,consumption,finance,medical care and so on.In this environment,as a kind of financial technology,P2 P automobile online loan has also risen rapidly.In the market formed by P2 P automobile online loan platform,how to reduce the risk of platform and fund has become a hot issue for scholars.A large number of personal and credit information submitted by the lender,as well as a lot of information about the loan products,are aggregated and manually audited to form a loan order.Such a huge amount of information invisibly increases the complexity of identifying the real situation of lenders and making lending decisions,but in the era of big data,the huge and high-dimensional data set formed by loan orders has also become an important tool for knowledge discovering through data mining technology.In order to reduce the risk of lending between the online lending platform and the financiers,and to reduce the complexity and labor cost of the online lending platform in screening the high-dimensional information of the lenders,the purpose of this paper is what kind of lenders can ultimately get the full approval of the platform loan or be rejected.In this paper,we use some loan order data of Meili Motor Finance Network Loan Platform from 2015 to 2018,introduce decision tree and random forest algorithm in classification and prediction model,use four algorithms to model,train and test the model,and finally compare the performance of each model,select the best algorithm to find and summarize the rules,and give scientific management suggestions.The results at the algorithm level show that,the model generated by CART algorithm has the best performance,but the performance of random forest model is better than that of single decision tree model.In the ranking of importance of attributes generated by Gini Index and accuracy,the top ten attributes of importance are whether to install GPS,whether to have life insurance,whether to have bank reserved mobile phone number,channel verification results,whether to have bank account number,whether to have ID card number,current position and type of work.It can be seen that the more open the borrower's information,the easier to obtain loan approvals.The higher the borrower's ability to consume,the easier it is to get loan approval.The first 5 and 10 important attributes are discovered by decision tree algorithm again,and the rules are as follows: when the lender does not install GPS,and there is no life insurance,and no repayment bank reserves mobile phone number,the final approval results are rejected.In addition,according to the mining rules,it is found that the platform verification means are relatively rigid and single,which may lead to customer loss.It is suggested to introduce ZHIMA CREDIT,Du Xiaoman Finacial and other third-party credit agencies as one of the verification means to objectively present personal credit.