Research On Application Of Combined Model In Online Lending's Anti-fraud

With the rise and prosperity of the online lending industry,many fraudsters have focused on this industry,and even formed a special black industry chain to defraud the loans from the online lending platform.Fraud in the application has become an inevitable pain for the online lending industry.The ability to control the risk of the online lending platform is an important factor determining whether it can develop healthily in the long run,and the ability to control the fraud in the application is even more important.In order to enrich the anti-fraud means of online lending platform,based on the experience of the predecessors,the theory of fraud scorecard of the GBDT-Logistic regression combined model is proposed and corresponding empirical research is carried out.The specific steps of fraud scorecard of the GBDT-Logistic regression combined model are as follows: Several original variables are used as the input variables of the GBDT model,and the indicator whether fraud happens is used as the dependent variable,and parameter selection and optimization are carried out.Then the fraud probability is generated by the optimized GBDT model.After WoE transformation,the fraud probability is used as independent variable and the rest of the selected original variables are transformed into the factor scores by factor analysis.The factor scores are input as the independent variables of the logistic regression model to get the logistic regression result.Then the factor terms are reduced to the equation of the input variables according to the coefficients of the logistic regression and result of the factor analysis.Finally,the scorecard is generated according to the score formula.In this paper,the fraud scorecard of the Logistic regression model,GBDT model,the fraud scorecard of the GBDT-Logistic regression combined model are empirically studied,and the results of these three models are compared.The comparison results show that the fraud scorecard of the GBDT-Logistic regression combined model is more stable and more interpretable than the GBDT model,and the fraud scorecard of the GBDT-Logistic regression combined model has stronger ability in distinguishing and ranking than the fraud scorecard of the Logistic regression.