The Research And Application Model About Churn Prediction For Mobile Customer

Using The churn prediction model for middle and high value mobile customer, the customers'characteristic and behavior can be analysed in a more detailed way, and customers who will churn recently can be more likely identified. There are two main advantages of precise marketing for high possible churn customer. Firstly, the marketing cost will be reduced greatly. Secondly, those customers who do not intend to leave may not be disturbed, which avoids making them angry. Also, building churn prediction model is really an effective measure to transform the CMCC Neimeng's management and marketing mode from extensive to precise.Some key issues about mobile customers churn prediction are discussed in this paper, such as defining time window, building validation dataset, filtering lost customers, and evaluating the defect caused by correlation between input variables and churn flag. Firstly, three decision tree algorithms, CART, C4.5 and CHAID, are used to build churn prediction model. Based on the comparison among Gini Index, information gain and Chi-square, the differences about precision and distinguish capability are analyzed for them. And massive experiment results also testify our conclusions. Moreover, we explain how to select data mining algorithm according to business requirement and data features. Finally, in order to understand this issue more deeply, we also build churn prediction model with Logistic regression model. In this model, PCA is used to and reduce the input variables and detail experimental results are listed.