Research On Single Credit Model And Integrated Model Of Personal Credit Risk Assessment Based On Data Mining

Personal credit is the basis of collective credit and national credit,on behalf of a person to comply with the extent of the agreement,affecting the community and others trust.The establishment of personal credit risk assessment system can reduce the credit risk of banks to a certain extent,and improve the competitiveness of the banking market to the national economic construction.In China,the establishment of personal credit system started in 2000,the typical method requires the applicant to fill out a form,according to the predefined scoring table score each indicator,the bank directly determine whether to accept the applicant's request.The score for each item in the scoring table is based on the subjective experience of the expert and is therefore not fair and reliable.In addition,if a certain score conditions need to be modified to adjust the score table is very troublesome.Therefore,to reduce the bank's credit risk,the development of personal credit business,we must scientifically and effectively assess the personal credit risk situation.The main work of this paper is as follows:The personal credit risk assessment model was established by using the statistical model(C4.5 decision tree,Naive Bayes)and the non-statistical model(SVM and BP neural network).They were compared and analyzed on three UCI datasets by using classification accuracy,Model stability and interpretability of the three models to evaluate the classification of several models.The experimental results show that the SVM model has the highest total classification precision(82.65%)and the second type has the lowest error rate(14%),but its stability is the worst.In addition to the stability,the BP neural network model is better than SVM(3.7%),and its stability is the best,but its classification accuracy is the lowest,and the error rate of the second type is the highest.The C4.5 model is the best in classification Accuracy and stability of the performance is not the best;for the interpretability,the statistical model has an absolute advantage.In summary,the existing credit risk assessment models have their own characteristics,but no one model allows each statistic to achieve a higher level.The homogenous parallel structure integration model(bagging integration and boosting integration)and the heterogeneous parallel structure integration model are constructed on the three credit datasets by using the integration theory.Experimental results show that the proposed method can effectively improve the classification accuracy and stability of the single classification model.For the homomorphic integration algorithm,the classification accuracy of the model after bagging integration is higher than that of boosting integration;the bagging integration and boosting integration model have similar stability effects;the boosting integrated model has lower error rate than the bagging integration.For the heterogeneous integration algorithm,the integration model is 4%higher than that of the single model;the integrated model improves the classification accuracy by 2%compared to the single model;the change of the integration model on the training set and test set(6.759%),which is close to that of the single classification model(7.24%).