Churn Prediction In Mobile Communication Network

In recent years, with the rapid growth of mobile users, mobile communications market is close to saturation. Customer churn has emerged as a critical issue for the telecom operators. Predicting churners from the Customer Relationship Management (CRM) data and Call Detailed Records (CDR) data capture many researchers’and telecom operators’attention. However, the problem of the loss of mobile users has not been solved either in theory or in reality. Therefore, it is of great practical significance and broad prospects to study the prediction of the loss of mobile users.Under the background in the era of big data mobile communication is coming, the mobile customer churn prediction was conducted using actual mobile users’data. And the performance of the prediction model is verified by large scale user data. In the research content, this paper carries out the research on the two aspects of the analysis of the characteristics of the loss of the mobile users and the prediction of the loss of the mobile users.On the analysis of the characteristics of the loss of mobile users, this paper focuses on the analysis of the relevance of the user’s voice business. Based on results obtained by the Modified Allan Variance (MAVAR), we confirm the long-range dependency for all base stations. But through comprehensive chi-square test we believe that the call arrivals can NOT always be modeled as Poisson distribution in short-term in most cases. We further analyzed whether user behavior in the voice business meets heterogeneous Poisson process characteristics from the three major characteristics of the heterogeneous Poisson process. We found that the characteristics of user voice service behavior if different at different time scales, further verify the user voice service correlation from the analysis results.In the prediction of churn of customers, the Artificial Neural Networks (ANN) are applied in customer churn as predictors. Moreover, we select input features based on the new features involving traditional demographic profiles, customer behavior information, as well as features selected by user correlation. Since we have illustrated the time-correlation of customers whose calls come from the same station. We established churn prediction model of the analysis of users’churn behavior in 45 days. Customer churn prediction performance was improved by adding user correlation feature. We analyzed the influence of features we selected on churn prediction through the analysis of the importance of feature analysis.In this paper, the relevant conclusions of the analysis of the characteristics of the loss of the mobile users can enhance people’s awareness of the voice behavior of mobile communication network users. At the same time, the results of user correlation analysis can be applied to the prediction of the loss of the user. The customer churn prediction model we established have important guiding significance for operators to predict the loss, and to take the user retention measures.