The Study Of Customer Churn Prediction Based On Data Mining

In the telecommunication industry, the competition for customer is more andmore fiercely between the carriers. As the market expansion become more and moredifficult，it requires carriers pay more attention on the maintaining existing customers.How to low down the customer churn rate should put the core position of thecustomer relationship management. So, more and more enterprises begin to analysiscustomer behavior features in-depth by certain technical methods. The main purposeis to establish customer churn prediction model and by this way to predict customers’turnover intention. It proves that the method has achieved good results. Data miningas an advanced algorithm and technology is applied more and more widely, and itplays an important role in areas of customer churn prediction.The management for churn of customers belongs to the category of customerrelationship management. With management concept, market situation and thepeople’s consumption concept changed, customer relationship management drawsmore and more attention by enterprises. To forecast customer’s turnover tendency isadvantageous for the enterprise to have time to respond to changes and to implementtargeted measures. In addition, to summary problems which lead to the loss ofcustomers in time may help enterprises to find their own shortcomings. Therefore, theestablishment of the customer churn prediction model warning mechanism has animportant sense for the enterprise. The main objective of this paper is to establish amobile customer churn prediction model.In the process of research, this paper collects theoretical knowledge on thecustomer relationship management and data mining, and makes full theoreticalpreparations for subsequent empirical studies. These theories involved the customerrelationship management of Customer Churn Management and customer valueanalysis; in additional also involved data mining theory. In the chapter it introducedthe commonly algorithms used in customer churn prediction in details; then itdescribes the model framework, which made the customer samples as the researchobject first, and after the data preprocessing it using two methods to establishprediction model for the customer data: the one is directly using the logisticregression model, the other takes the customer value into account. The main step isusing RFM model to distinguish customer with different value, and then using the K-means clustering method to make customer classification; finally, using the logisticmodel for each customer cluster and get some prediction models. At the end of thearticle we get some Tangshan city mobile company’s actual customer data forempirical research. The results show that compared with churn prediction modelwhich no consideration of the customer value, the churn prediction model consideredthe customer value has a high accuracy rate.In this paper, it establishes customer churn prediction model combined theoryand practice. The model is scientific and maneuverability, and is able to achievehigher prediction level.