Credit Risk Assessment Method Based On Random Forest In P2P Lending

In recent year,with the rapid development of P2 P lending,it has become one of the import financial industries.P2 P lending brings many opportunities for economic development.While P2 P lending brings convenience to personal financing,the high default rate is still a serious problem,which hinders the development of P2 P lending.Therefore,credit risk assessment has attracted the attention of scholars and enterprises in P2 P lending.With the arrival of the artificial intelligence wave,credit risk assessment methods based on machine learning has been widely used in P2 P lending research,due to its accurate predictive performance.However,the existing researches still have some shortcomings.On the one hand,the machine learning methods that aim to minimize default rate or maximize predictive accuracy,cannot guarantee investors' profit in the P2 P lending investments.On the other hand,compared with traditional credit risk assessment methods such as logistic regression and Score Card,the predictive results of machine learning methods are not fully trusted by the investors and P2 P platforms,due to that they are not interpretability.To overcome the above problems,this paper proposes the corresponding solutions.(1)In Section 3,the Random Forest optimized by genetic algorithm with Profit score(RFoGAPS)is proposed to evaluate the credit risk of the loan in P2 P lending.First,considering the actual and potential returns and losses,a new Profit score is proposed and taken as the optimization objective.Second,the genetic algorithm is used to optimize the combination of decision trees in Random Forest.Then,the dataset of Lending Club is used to evaluate the proposed method.Experimental results show that the RFoGAPS can obtain higher profits for lenders compared with actual profit and traditional methods.(2)To improve the interpretability of credit risk assessment method based on machine learning in P2 P lending,an improved pedagogical method for interpretable credit risk assessment is proposed in Section 4.Based on the traditional pedagogical method,the improved method adopts the Weight-SMOTE-based pseudo-dataset sampling method to facilitate the learning ability of the decision tree for the correct and high-value mapping relationship in the underlying black box model.In addition,to overcome the shortcomings of the fidelity,a novel evaluation method,named by actual fidelity,is proposed,which can effectively measure the learning and simulation ability of the decision tree for the correct mapping relationship in the underlying black box model.In the end,our method is evaluated in the P2 P lending data set,and the experiment results show that the proposed method can effectively assist investors and platforms to interpret the black box model of credit risk assessment in the P2 P lending.The research on the credit risk assessment method based on Random Forest can further enrich the theoretical system of credit risk evaluation in P2 P lending,and promote the application of machine learning method in the credit risk assessment of P2 P lending,which has a good practical application prospect.