Companies in the Internet age are faced with a big problem calling “Customer churn”. Research has proved that an old customer produces more returns than a new one. At the same time, attracting a new customer costs more than keeping a losing old one, which makes it important to build a model to predict the customer churn. Firstly, this paper will introduce how to choose the model data and the definition of the variables. Then, it will select the right variables by using bivariate analysis. Next, this paper respectively use the method of dummy variables and the method of WoE, according to the users’ behavior characteristic variables, to establish logistic regression model, classifying and grading the tendency to lose, predicting the possibility of the lower consumption in future time. At the same time, this article will also test multiple collinearity and variable correlation of the model, making the model more comprehensive. At last, it will compare accuracy and stability of the two methods through the model results and KS value. |