Research On The Personal Credit Score Of P2P Platform Based On Decision Tree C5.0

As an important component of inclusive finance,P2P lending plays an important role in personal loans and financing of small and micro enterprises by virtue of its convenience,efficiency and low investment threshold.But everything has two sides,the P2P network to borrow at the same time of rapid development,is also facing a significant credit risk,especially in the present regulation tightening,break the rigid to honour a trend of real cases,how to control from the source the borrower credit risk becomes very important for P2P.Personal credit risk score is effective to identify and reduce the credit risk of the decisive factor,and the decision tree classification model,relying on its advantages of high classification precision and easy to implement by application in the field of personal credit score.This paper selects the loan data of pure credit target of domestic online loan platform as the research object,and use the latest decision tree classification model C5.0 algorithm to predict and evaluate the credit risk of borrowers.In actual score,there is the so-called "garbage in,garbage out",namely the score data affect the result of the score predicts significant,article 5C+1S method based on network platform loan borrowers data classification,build credit rating index system of the properties for the category fields to merge,and the reasonable value of the category field dimension,in order to achieve smaller model and prevent excessive fitting effect.The empirical results show that the prediction accuracy of training sample classification after data processing is reduced to less than 1%,while the prediction accuracy of test sample classification is increased by 3.16%.In model optimization improvement,the article introduces miscalculation cost and misjudgment matrix,by raising the second wrong misjudgment costs,achieve the purpose of reduce the second category of error rate,it also conforms to the actual score in the work details.Empirical study shows that after the miscalculation cost matrix to determine reasonable,although the test sample set the first type of error rate increased by 2.93%,but the impact on the actual score loss bigger fell by 7.35%in the second category of error rate.According to the final decision tree C5.0 model,the conclusions of the importance of each variable are outputted.The P2P Taipei should be in at the same time,strengthen their own credit risk control ability will docking central bank credit credit reasonable combination,with a third party and use of big data inquiry expand dimensions and depth of the borrower credit checks.