Application Of Logistic Regression In Personal Credit Rating

With the rapid development of economy and the progress of Internet financial technology,the domestic microfinance business has provided a fertile environment for small loan companies.From 2006 to 2015,microfinance companies enter the crazy growth pattern,in the past two years some indicators growth rate slowed down,tend to be rational development.Nowadays,some Machine Leaning methods are widely applied in personal credit ranking,such as Logistic Regression,Discriminant Analysis,Decision Tree and Neural Network.This paper apply Logistic regression to establish a personal credit ranking model which is the most widely used in practice to quantitative the default risk of microcredit customers,in this way,achieving the goal of risk control.Considering the logistic regression model during the learning process,the amount of information between default customers and normal customers,this paper use SMOTE algorithm to generate synthetic default data,which could enhance credit ranking model's TNR and overcome the different information amount between small class and large class in sample,and keep information the same between default customers and regular customers.The discretization technique are applied in the logistic regression model,logistic regression model belongs to the generalized linear model which expression ability is limited,the use of entropy-based data discretization method can significantly improve the stability and interpretability of the model.In this paper,unprocessed by discretization and SMOTE credit rating model's accuracy is 81%,TNR is 57.77%,AUC is 0.8283.Contrast with the logistic regression credit rating model processed by discretization and SMOTE,the model's accuracy improved to 85%,TNR is 83.33%,AUC is 0.89.The predict accuracy and goodness of fit of the model is significantly improved.