Customer Churn Prediction And Analysis Based On Telecom Social Network

With the rapid development of mobile communication and Internet industry, the competition among domestic operators is becoming more and more intense. The replacement of operators by telecom users is becoming more and more common, and the phenomenon of customer churn has attracted wide attention in the industry. How telecommunication operators can give full play to their advantages in order to achieve a higher customer retention rate, and even attract more new customers to join the company, will directly affect its position in the telecommunications industry market.This thesis researches the current works on the customer churn prediction. Different from the traditional research emphasis, this thesis focuses on the prediction of the early churn prediction, which means predicting customers' churn possibility some months later except for the trend of churn in net month. Based on the tendency of customer features along with time, this thesis analyzes users' time series made of feature set during several months, and validates the feasibility of early churn prediction based on time series. Then, three early churn prediction models based on time series are constructed, as well as a prediction model based on social influence from churners in the network. The four models are merged to obtain the final mixed model.This thesis also analyzes the feature importance of the customer churn prediction, trying to find the early warning signs for churners. Also we proposes a re-entering customers identifying model based on social similarity, which will help operators to develop accurate marketing strategy for such users.Based on the distributed system, we implement the models separately, and the results of the models are compared with the user's call record and ServInfo data provided by China Telecom. The reliability of the proposed models are verified. Based on the models, this thesis constructs a telecom user churn prediction and analysis system based on Spark platform, and integrates the early churn prediction and the re-entering customer identification function into the system. User can complete the model training applying automatically, and get the corresponding potential churners and re-entering users.This thesis completed the early churn prediction and analysis, as well as the re-entering user identification. Also this thesis constructed a complete customer churn prediction and analysis system. According to user's need, the system complements automation user identification and prediction work, which is critical for telecom operators to develop targeted marketing strategies.