Customer Churn Prediction Of Mobilcom Based On Fuzzy Bayesian Network

The number of customers is essential for any company. So how to tap new customers on the basis of resources, and protect the original client resources are very important and no enterprise can ignore this problem. Especially in the mobile communications industry, due to customer volume is fixed, the services provided by various companies are similar, how to maintain and expand existing customer base become key point. Currently, many mobile communication companies do not have effective customer care systems. When they spend a lot of resource of digging new customers the loss of existing customers is increasing. As a result, not only customer base has not expanded, but also resulted in an increase in operating costs and a large number waste of resources.To solve the loss of the mobile communications industry clients, this dissertation study a particular mobile communication company. Through the analysis of data of the company, we found the number of customers is huge, and the data of the customers are complicated. The company takes no use of these data to establish effective, predictive mechanism to support decision-making. Therefore, we can see the loss of some customers but do not know what kind of customers need to maintain in order to prevent further loss. Based on the above considerations, this dissertation summarizes three categories, eight small reasons of the customer loss, and analyzed three-point problems which can be improved on the company level. And then reorganized the basic information provided by customers information in order extracted out of eight pairs of critical points which makes a serious impact on the customer loss. By converting the impact of customer loss which affiliate with people's satisfaction into a precise probability, build bridges between people's perception of the product and customer loyalty. Through combining the quantitative data and qualitative people feeling, set up a fuzzy Bayesian network. Then using historical data to learn and correct parameters, in the end calculate and predict customer's future loss possibility. Comparing the result with the realistic data in the test pool, the error is less than 10%.The experimental data have proved the model in predicting the loss of future customers have good availability and high accuracy.