Loss Prediction Based On Support Vector Machine Mobile Telecommunications Industry Clients

At present, as the reformation of telecom industry on our country, the competition among every domestic telecom operators is becoming more and more drastic in the field of customer service. The drastic competition leads subscribers to stay in unstable state. All the operators are facing a problem of serious customer churn .With the coming of foreign telecom operators, this situation cannot but becoming worse and worse .The frequent custom churn delays the reclaim cycle of company's profit, and brings tremendous loss to the enterprise .In this situation .How to predict customer churn according to the customers' character and behavior habit ,and draw effective strategies to prevent them from leaving so as to produce profit, is an important task to keep the telecom company successfully running ,and it is also an important problem in the field of customer research and consultation.To resolve the problem of customer churn in the mobile telecom company, this paper brings forward a feasible method, that is to set up customer churn model according to the existing information system, especially the Union Business Information System Platform (UBISP), combining the knowledge of data mining, and forecast the potential lost customers, then take pertinent measures to prevent them from leaving according to their character of calling and their service fancy. This paper introduces the design steps of customer churn predictive model. And it emphasizes some questions in the model designing, such as data distilling, the pretreatment, model validating.In this paper, a method of applying SVM into the customer churn predictive model of the mobile telecom is put forward, according to the actual state of application of data warehouse and data mining in the mobile telecom company. The relevant theory about data warehouse and data mining, is expatiated further in the paper. Especially, as a new method of data mining, the theory of support vector machine is systemically introduced.In the end of the paper, a simulative research on the customer churn model based on the method of SVM is carried out. The process of simulation includes two steps. The first step is finding the predictive model through training and learning from customers' data which is known by us now. The next step is forecast that which customer is likely toleaving the company, by using the model. After the whole process of data mining, a satisfactory result comes up. Some primary research about the affection to the result with different kernel functions and different parameter is done at last.The mobile telecom customer churn prediction scheme based on SVM in this paper is hoped to provide some references to the resolving of the problem of customer churn in our country's telecom industry.