Analysis On The Factors Regarding Telecom Broadband Customers Lost

Now days, with the introduction of competition in the telecom market, there are more choices for the customers in terms of the products and the service provider, and the scramble for customers among those telecom service providers are becoming tough consequently. Meanwhile, the traditional network and technology advantages are fading and there is no gap among the telecom service providers, no one can lead the industry by building the technical gap advantage. Therefore, to create the new leading advantage for competition in such a new situation, all the telecom service provides are aiming to focus on the customer, know the customer deeply in all aspects, guide the customer and hold them.This research presents the purpose on a discussion about the main factors that influence the telecom customer lost. It also set up the mode for the lost prediction. With the application of customer category, service quality, customer satisfaction and loyalty, the basic mode for the customer lost is set up, and in addition, 20 variables which have their effects on the customer lost are found.This research build up a process for how to construct the prediction mode for telecom customer lost. It concentrates on the ADSL broadband individual users of A company, obtains the material via data collection,finally, 20 variables which related to the customer lost prediction are summarized and served as the analysis objects through data digging method. After the sample data cleaning and coding, they will be studied by using the basic statistic method for the distribution of each variable.5 variables without significant effects would be discarded by the Logistic regression analysis, the other 15 remained and discussed by using the decision tree analysis subsequently. Based on the importance of attribute method, they will be divided into the number variable and class variable for prediction power inspection. The 15 variables are sorted by the prediction power and verified by the category mode, and then the positive relation threshold between each prediction power and the overall prediction power of the model can be found. For those unstable variables, they will be evaluated respectively to see whether it contributes to the overall prediction or not, and kept or discarded accordingly. Eventually, 12 variables which have strong prediction power on the customer lost are reserved to set up the prediction model for customer lost, which satisfied well for A company's ADSL broadband individual customer lost.