Research And Application Of Telecom Customer Churn Prediction Based On Fuzzy Bayesian Network

In recent years, the domestic mobile Internet had a strong development, especially the smart phones. The world had entered the mobile Internet age. It had evolved into a new way of life that information exchanged by using mobile terminals to access the Internet. It had a report from China Internet Network Information Center (CNNIC) in2014, showed that Chinese netizen had reached618million at the end of2013and the Internet penetration rate was45.8%. Among them, the mobile phone users had reached500million, its annual growth was19.1%.In addition, because of the products and services provided by operators are similar, the three operators have changed their market strategy:from "product-center" to "customer-center". This leads that customers become an important resource to them. Who has more customers, whose market share is bigger. So they begin to take various measures to attract new customers. Then, the existing customers are ignored. With the growing churn rate, especially the high-values customers, there exists a scene:no increase in increments. As a result, an important issue needs to be solved, which is to predict customer churn accurately and effectively and take scientific and feasible measures to retain and reduce churn rate.There is mass communication data produced everyday. They occupy a lot of storage space and also imply much commercial value information. If those data can be mining, it will generate a great wealth.Based on above backgrounds and development requirements, this paper analyzes customers’real consumption data provided by one operator, and launches the following studies:1. After the current telecom operations are analyzed, the various phenomena and the cases of the telecom customer churn are deeply studied. The current studies in the telecom field and the applications of the data mining are discussed and overviewed. 2. The applications of the Bayesian network and fuzzy theory are deeply discussed in the telecommunications business data, a hybrid telecom customer churn model is designed and implemented based on fuzzy Bayesian network.3. The structure learning and parameter learning algorithms of Bayesian network are studied. The method converted from degree of membership to fuzzy probability, as well as the calculation method of fuzzy joint probability and fuzzy conditional probability, is designed and implemented.4. After the reasons of the customer churn in the telecom company is analyzed, this paper analyzes the reasons of the customer churn, the churn-related characteristics and data attributes are extracted from all attributes of the data set. The samples are obtained.5. Through experiments of sample data, the effectiveness of applying fuzzy Bayesian networks in the telecom customer churn is verified in this paper.At last, the work of this paper is summarized and the direction of the future work is given.