Research On Distributed Customer Churn Prediction Model Based On SVMs

The global financial crisis spreads to the retail industry and causes more competition. It is no doubt that the retail companies should maintain their customers through customer churn prediction in order to survive in crisis. In this case, how to analyze customers' consumption behavior and find the potential rules from the massive data by using data mining technology become more and more important. We should focus on enhancing customer satisfaction, loyalty and reduce the rate of customer churn for surviving the crisis successfully.The chain-store operations in retail industry are well-developed in a high speed through mergers and acquisitions. The database develops from centralized management to distributed management. A large number of chain stores, distribution centers and headquarters have constructed commercial data-sharing environment through network interconnection. The traditional method of customers churn prediction could not efficiently make a global decision-making under the distributed environment. On another hand, Distributed Data Mining technology can analyze and discover comprehensive business information from distributed databases. It provides a new means for customer churn prediction at present.Based on the existing research, this paper builds a Distributed Customer Churn Prediction Model in Retail Based on SVMs named R-DCCPS. The outline contents in this paper describes as follows:Firstly, after studying the current situation of customer churn both at home and abroad, then do research on the new features came form the chain retail industry now. The dimensions of customer churn analysis were introduced by this paper, including customers' behaviors on customer satisfaction, customer loyalty, and purchase frequency, brand transfer and customer value. It set an index system of customer churn in chain retail by analyzing muti-dimesition impact factors.Secondly, the Distributed Customer Churn Prediction Model in Retail Based on SVMs named R-DCCPS is proposed, which takes the customer data in various distributed nodes as the data source, takes mobile agent operation platform named Gmine system as the framework. It uses PCA and SVM as the key technology to implement the high performance Algorithm. With Eigenvalues Tree carried the information of support vectors, it can get global knowledge from distributed databases for realizing the efficient and accurate business decision-making in chain retail.Thirdly, Because of the distributed and heterogeneous data, a new data storage structure named Eigenvalues Tree was introduced in this paper. It uses the eigenvalues tree modified by SVDISVMs and HVDISVMs algorithm as the middle process, gains the integrated Support Vectors from distributed business database and uses this Support Vectors to construct the global classifier plane for realizing the customers churn prediction.Fourthly, a data mining prototype system is designed and implemented for customer churn prediction of chain retail enterprise. A specific business application with model and algorithm outlined above is used in Gmine system, which has a good scalability. It has found that the system improved the chain retail enterprise's decision analysis and support, and gave a strong support to CRM management.