Application Of Random Forest In Credit Risk Evaluation Of P2P Lending

In recent years,with the gradual development of the Internet Finance,P2P lending has become a new popular pattern of Internet Finance nowadays.With the advantage of high yield,various P2P lending online platform have showed explosive growth.However,due to the later start of domestic P2P lending,the imperfect credit system and the lack of relevant laws and regulations,which making investors face a more serious security problem.Credit risk also has become the main bottleneck in the development of P2P lending industry,it's very urgent and necessary to establish a good risk assessment model for P2P network borrowers.While it has lacked systematic and in-depth research in the academic community,risk assessment is still in a stage of simple imitation of the traditional personal risk assessment methods.In this article,we choose a model of combinatorial classifier-Random Forest,which has better toleration of noise,difficulty to be over-fitting,and a high stability.Compared to the traditional model of single classifier,it can be better to deal with credit risk assessment problems.In this paper,the theory of random forest algorithm is introduced in detail,and the weighted random forest algorithm(WRF)is proposed based on this model by introducing cost-sensitive learning method.Thus we can improve the accuracy of mistake cost higher category,and enhance the practicability of model.In the Empirical research,Firstly,the data is preprocessed,including getting rid of the outliers,polishing the missing values,the normalization and correlation test.Secondly,I use the RF algorithm to determine the selection of indicators with the five cross data.Thirdly,I establish a personal forest credit assessment model based on random forest with the Lending Club open data set,German and Austria Credit Data Set.And then compare with the traditional credit risk assessment methods:Logistic regression,KNN,SVM,ANN.By comparison,we find that the RF model has a higher overall accuracy,which show that the RF algorithm is more suitable for building a credit assessment model.Finally,balancing the P2P credit data with the SMOTE algorithm,we find that the classification result is more realistic.