The Research Of Low-end Customer Churn Prediction Based On Cox Model In Mobile Telecommunication Industry

With the development of telecommunication industry, to maintain customers is becoming the focus of domestic major telecommunication operators, as the great loss results from a large number of customers churn. Aimed at the customer churn problem, This dissertation proposes that the Cox model is applied to customer churn prediction. By studying the historical data provided by a subsidiary of China Mobile Telecommunication, we identify the attributes that significantly affect customer churn and then develop a prediction model using Cox's proportional hazard model.This dissertation analyzes various factors that may cause customer attrition, examines their correlation with churn behavior, and illustrates the problems with variable analysis. Then, significant predictors are selected by multivariable analysis, a customer churn prediction model is built using Cox's proportion hazard model, after checking the basic assumption of proportional hazards. At last, the model is valuated using the hold-out sample.The empirical results suggest that the significant prediction variables include : age, whether the customer is local resident, whether the customer leave phone number, how much message-fee per month, the proportion of call time between different networks, how much long-distance call fee, and so on. Use of these co-variables, through the training samples to establish Cox proportional hazards models, at the same time, the prediction of validation sample 86,330 (including 79042 in the net, off-grid 7287), each customer's probability of off-grid can be clearly appear, and then take off-grid probability of descending sort, which in the front 10 percent of the customers 8663, including the loss of customers 6517, accounting for 89.43% of the total loss. In other words, as long as the customers that are belonged to the front of 10% can be retained, 89.43 percent of the loss of customers can be retained.It's also suggests that this method has a good effect on customer churn prediction for the mobile communications industry, Enterprises can significantly reduce the retain cost.