Research On Prediction Model Of Telecom Customers Churn

With the deepening reform of telecommunication systems, the competition betweenoperators of domestic telecommunications industry becomes increasingly fierce. Meanwhile,operators tend to follow a variety of promotional activities and an endless stream ofadvertising to attract new customers in order to gain more customers and take up a largershare of the market. However, statistics found that the development of a new customer costs7times higher than to keep an old customer, and if “user retention rates” increase5%,85%profit growth is expected to bring to operators. So old customers retained is directly related tothe interests of operators, both customer loss and the loss of traffic volume will have aprofound impact on operators. For this problem, the paper studies characteristics of losingcustomers with data mining, and thus to predict the loss and assess the consequences of theloss, customer retention measures taken to prevent leading management crisis, and also toenhance the competitiveness of telecommunications companies.The main fields of application research of data mining in the customer relationshipmanagement in the telecommunication industry and the related application are studied in thispaper, and the causes of customer churn are also analyzed in order to differentiate betweenlosing customers, thus leading to wastage of different standards. There prediction algorithmsare decision trees, neural networks, support vector machines and logistic regression, etc. butall of them have advantages and disadvantages. C5.0decision tree, C&T decision trees, neuralnetworks and support vector machines are used in this paper and results show that C5.0decision tree has the highest resolution accuracy rate of90%, followed by SVM reached morethan80%. In order to increase the models’ feasibility and accuracy of prediction for differentdata, a comprehensive model is put forward, which is achieved by adding a confidenceinterval above the models for integrated. Finally, the integrated model is built. This proposedmodel combines the advantages of a variety of forecasting methods, improving the accuracyof forecasts and reducing the risk prediction, and it provides a theoretical basis for loss forecast algorithm.