Research On Customer Relationship Management Of Mobile:Based On Data Mining

Accompanied by multiplying information,data mining is an emerging subject intercrossed which used as method for the discovery of knowledge. With the developing of datahouse, data mining has appealed much attention. It has spread extensively since the technology has being perfected in algorithm. As to the telecommunications industry, it has hundreds of millions of information about it’s clients, which are recorded by clients every time they use it. For such a large amount of information,it is a effective means to explore the right information by the application of data mining.Nowadays,with the product getting increasingly identical, the operators of telecommunication make the quality of service as important means to improve the core competitiveness by tackling with homogenization. Quality of service made the operators pay more attention to customer relationship management,which is a new management pattern. Based on customer segmentation,the competitiveness of enterprises enhanced by applying of personalized and different services. The operators attach high attention of how to explore the clients with high value in innumerous clients, for the profit level is determined by the quality of it’s customers directly. The customer churn in customer relationship management has becoming an urgent problem that the operator must take effective action by recognizing the customer churn when it tends to happen through customer churn prediction. It will surely avoid the profit loss and loss on image brought by customer churn.Based on the analysis of the customer relationship management,the article defined the connotation of it, and defined the mobile customer with high value from three valuable dimensions including the current, future and the nonmonetary one. Six indicators are used for the measurement of the current value, such as the daily rental, local telephone charges, long distance charges, roaming charges, charges for short message and GPRS; future value included five indicators:the flow, duration for roaming call, long distance and local call,as well as the amount of short message; the duration for charged call and client’s status are two indicators for the nonmonetary value. The application of K-Means, which is a technology among data mining,to pick up the right clients with high value by segmenting the mobile clients. From the result of the model,the right clients takes up one fourth of the whole, with each index an important one. Based on the related conception of customer churn, evaluating the pressure of customer churn of mobile customers, and analyzing the reasons of customer churn, the article set early warning models of customer churn for mobile clients by applying of C5.0decision tree model and B-P(Back-Propagation) neural network model. C5.0decision tree model’s key grouping variables are:duration time, daily rental,duration of long-distance call, their ages, and the type of chosen package. Model of C5.0is a decision tree of five layers,with prediction accuracy of98.22%for the training samples, and of98.21%for the test samples. The first five grouping variables of B-P neural network model are:duration of long-distance call, their ages, charged local calling, charged roaming and the type of chosen package. The structure of Neural network model includes11input nodes; a hidden layer, with20hidden nodes; and1output node,with prediction accuracy of98.25%for the training samples, and of98.17%for the test samples.The two has minor in difference from the view of prediction accuracy. But the C5.0decision tree model is more stable,if we take view from graphs of response, graphs of effectiveness and graphs of the ratio of investment and return. So we get the conclusion that it is ideal to predict the customer churn with the application of C5.0decision tree model.