Prediction And Evaluation Research Of Customer Losing About The Telecom Operation Companies

At present, the telecom operation enterprise customer loss is a complex probleminfluenced by various factors. Especially China’s telecom industry has conducted a newround of telecom restructuring again according to the issuance of3G licenses, since then theentire business operations under three operation enterprise have had a fierce customer marketcompetition. Because of the huge mobile customer group of China mobile and the strongmobility of the low-end customers between different operation enterprise, in view of thecustomer loss, analyzing and building customer loss prediction model has importanttheoretical value and practical significance.This paper analyzed the research of scholars in the field of customer churn in detail,discussed the factors and prediction methods influencing the customer loss. Through theanalysis of the telecom operating environment of3G era, the current situation of the domesticand foreign telecom operation enterprise customer loss were summarized. In the angle oftelecom operating environment and the operation enterprise losing customer data statisticalanalysis, the causes of the telecom operation enterprise customer churn were discussed deeply.Then, eight types of the customer loss causes were summarized. Accordingly, based on datamining and customer value theory and method, the application of the BP neural networkalgorithm, the algorithm of support vector machine, C5.0decision tree algorithm in customerloss prediction were studied. In order to obtain better forecast effect, Lagrange combinationforecast model and prediction model based on customer value were constructed. We havefocused on the following questions:First, based on studying and drawing the domestic and international relevant data miningtheory and results extensively, we explored the customer structure of the telecom operationenterprise, analyzed the concept of customer loss and losing customer, as well as thephenomenon and characteristics of customer loss, thus combing three relation models havebeen organized.Second, carrying on the constructing model of customer attribute classification study, weput forward primary attributes and derivative attributes’ effect on studying customer lossprediction comparatively. In the past, the studying on telecom customer loss prediction was onthe basis of customer consumption behavior, personal information, payment information and other original attribute data which are hard to truly reflect the behavior of the customer loss;The derivative attributes were added, such as: rent signs, call turn signs, the account balancesigns, recharge behavior signs and so on. The data set can better predict the customer loss,making prediction hit rate higher, the research significance of customer value calculationmore.Third, by analyzing customer agreement data，consumer behavior data and billing data，attribute set relating the customer loss were concluded. According to the ease of access to theoperating enterprise data, the customer loss prediction index system was established. At thesame time, the Lagrange combination forecast model was built based on data miningalgorithms. For customer loss prediction research, three classic data mining algorithm (BP,SVM, C5.0) were selected to build a single customer churn prediction model, and the modelassessment shows that not any single model are optimal. Accordingly, with the thought ofLagrange function for extreme, the combination forecast model of customer loss isconstructed whose forecast effect is more ideal than a single model.Fourth, that whether the customer loss list resulting from he prediction of thecombination forecasting model is valuable to retain or is it necessary to reinvest for thesecustomer is depending on if they are valuable for customers to the operating enterprise.Therefore, the methods based on two dimensions to enhance customer loss prediction resultswere proposed. The two dimensions are customer loss combination forecast based on theLagrange and the loss prediction on the basis of customer value. Then according to theprediction results of these two approaches, the root causes of customer loss have beenanalyzed.Finally, through the analysis of the causes of the customer loss, as well as the assessmentof customer churn prediction model, measures and suggestions to reduce customer loss for thetelecom operating enterprise would be proposed.