Overdue Forecast Of Microfinance Users Based On Random Forest -Logistic Regression Combined Model

Personal credit is the cornerstone of the entire social credit.In market transactions,all economic activities are closely related to personal credit.Therefore,how to correctly evaluate personal credit has an important significance.However,with the development of the economy,on the one hand,the contradiction between the increasingly important credit record and the lack of credit record has become increasingly intensified.On the other hand,in recent years,with the continuous development of individual-oriented micro-finance business,preventing personal credit fraud and reducing the rate of nonperforming loans have become the primary objectives of related business,so it is urgent to establish an appropriate personal credit evaluation model.In other words,the ability of financial institutions to prevent fraud and reduce the defective rate is the core of risk control for traditional financial institutions such as commercial banks and micro-finance corporations.It is also the key to the sustainable development of emerging Internet financial institutions such as P2 P.This paper uses 11017 real data and 199 credit characteristics provided by UnionPay Merchant Services after desensitization.First of all,we should preprocess the data and filter the characteristic variables,then theoretical analysis of the existing single model.So we selected random forest model and logistic regression model to combine from the prediction accuracy,robustness and model explicable.The output of the random forest model and the selected characteristic variables are recombined as input variables of the logistic regression model.A two-stage credit evaluation model of random forest-Logistic regression is formed.Then,the accuracy of the combined model is calculated according to the comparison between the predicted results and the real value and compared with the prediction accuracy of two single models of random forest and Logistic regression.