Font Size: a A A

Research Of The Model And Arithmetic Of Customer Churn Prediction Based On Business Intelligence

Posted on:2008-05-17Degree:DoctorType:Dissertation
Country:ChinaCandidate:G E XiaFull Text:PDF
GTID:1119360215959148Subject:Management Science and Engineering
Abstract/Summary:PDF Full Text Request
Customer churn management is important for many industries. In recent years, the researches on customer churn prediction have came out with many great results on traditional statistics methods and artificial intelligence methods. However, there are still many fields need to be studied. The evolution of business intelligence(BI) provides a new way for studying the customer churn prediction.This paper revises current model structure of customer churn prediction and establishes a new model structure of customer churn prediction under the frame of BI and customer relationship management (CRM)theory. Based on the new model structure,customer churn management strategy model, feature extraction, attribute selection and prediction model are proposed. This paper investigates empirically customer churn prediction for the telecommunication industry. From a new point of view, it reinforces the comprehension of customer churn rules. Finally, the strategies are studied for controlling customer churn. The concrete researches include below content.1,A new structure for predicting customer churn is proposed and consequently a fire-new thinking is introduced to solve the difficult problem of customer churn prediction. According to the characteristics of customer data and the drawbacks of the existing methods, this dissertation uses a new way, including feature extraction, attribute selection,prediction model design to solve the difficult issues of customer churn. Experiments validate that the new model structure is more effective than the current model structure.2,A customer churn management strategy model(CMSM) is proposed for the telecommunication industry, which is based on DELTA strategy model for firm competitiveness。CMSM is substantiated by using a customer churn dataset of a telecommunication carrier. It is found that the model could describe customer churn reasons and include the main predictor factors related with firm competitiveness strategy, thereby its application is easily controlled.3,The paper empirically analyzes the customer churn prediction in the industry by using feature extraction and attribute selection methods. Main conclusions lie in four aspects.Firstly, KPCA is introduced into customer churn prediction, and the corresponding feature abstraction method is presented. The prediction model is designed by combining KPCA and Logistic regression. Experimental results of customer churn prediction for a telecommunication carrier show that the proposed method is superior to original attributes and PCA feature abstraction in hit rate,covering rate, accuracy rate, lift coefficient, hit rate confidence interval, covering rate confidence interval, accuracy rate confidence interval and Kappa, which indicates that KPCA abstracts nonlinear features of customer data and provides an effective measurement for customer churn prediction.Secondly, Information gain(IG) is introduced into customer churn prediction, and the corresponding attribute selection method is presented. Information gain neural network(IGNN) model is designed by combining IG and NN. Experimental results of customer churn prediction for a telecommunication carrier show that the proposed method is superior to correlation selection,consistency selection, instance -based selection and symmetric uncertainty selection in hit rate, covering rate, accuracy rate, lift coefficient, hit rate confidence interval, covering rate confidence interval, accuracy rate confidence interval and Kappa, which indicates that IGNN has stronger capabilities of prediction and generalization than NN to be expectantly applied to practice. Hence,the validity, credibility and reliability of the proposed method are verified.Thirdly, Attribute selection is essentially a satisfactory optimization problem in customer churn prediction. Most of the existing attribute selection methods did not consider the cost of attribute extraction and automatic decision of the dimension of attribute subset. In this paper, a novel approach called satisfactory attribute selection method (SASM) is proposed. SASM considers compromisingly classification performance of attribute samples, the dimension of attribute set and the complexity of attribute extraction. Attribute satisfactory rate and attribute set satisfactory rate are defined. Several satisfactory rate functions are designed. Satisfactory attribute set evaluation criterion is given in a mathematical way. Satisfactory attribute selection algorithm is described in detail.Experimental results of customer churn prediction for a telecommunication carrier show that SASM is superior to correlation selection, consistency selection, instance-based selection and symmetric uncertainty selection in hit rate, covering rate, accuracy rate, lift coefficient, hit rate confiden-nce interval, covering rate confidence interval, accuracy rate confidence interval and Kappa. Hence,the validity , reliability and applicability of the proposed method are verified.Fourthly, with the increment of time space, the prediction model based on feature extraction and attribute selection should be trained over again to acquire the satisfied results.4,The paper empirically analyzes the customer churn prediction in the industry by support vector machine on structural risk minimization. Main conclusions lie in five aspects.Firstly, This paper studies customer churn in telecommunication industry by using normal SVM. The method is compared with aritifical neural network, decision tree, logistic regression and naive bayesian classifier regarding customer churn prediction for a telecommunication carrier. It is found that the method has the best accuracy rate, hit rate,covering rate, lift coefficient, hit rate confidence interval, covering rate confidence interval,accuracy rate confidence interval and Kappa , except that C4.5 is superior to the proposed method in hit rate and hit rate confidence interval. It provides an effective measurement for customer churn prediction.Secondly, Aimed at the shortcomings of the methods for customer churn prediction, the improved C support vector classifier(SVC) is developed by using the ratio of different classes in training set to evaluate the penalty parameters of the classes. compared with normal C-SVC, aritifical neural network, decision tree, logistic regression, naive bayesian classifier etc by predicting customer churn for a telecommunication carrier, it is pointed that the method can acquire the better accurate rate, in hit rate, covering rate, lift coefficient, hit rate confidence interval, covering rate confidence interval, accuracy rate confidence interval and Kappa, except that aritifical neural network, decision tree, logistic regression, naive bayesian classifier are superior to the proposed method in hit rate and hit rate confidence interval. The results indicate that the method could be an effective measurement for customer churn prediction.Thirdly, the customer churn prediction model by using simple SVM need less time for computing in the equivalent condition of model evaluation results.Fourthly, In this paper, a model based on SVM is used to predict customer churn. Through the analysis on the different effects of two types of errors in customer churn prediction, the tradeoff problem of two types of errors for customer churn prediction of a telecommunication carrier are is studied. Empirical results show that the model is efficient in reducing expectation loss function value on controlling the tradeoff problem of two types of errors by adjusting the "loss-ratio-coefficient" . It reflects the essence of the problem and can be a powerful decision-aided tool for customer churn.Fifthly, with the increment of time space, the prediction model based SVM should be trained over again to acquire the satisfied results.5,The classification table of customer churn is proposed for understanding customer value and customer satisfaction on CRM theory. Then, it is found that the thrust force and gravitation force lead to customer churn by analyzing the forces which effect customer churn. Finally, customer churn is controlled by making retaining strategies and resistance strategies.
Keywords/Search Tags:customer churn, business intelligence, prediction, statistical learning theory, support vector machine, feature extraction, attribute selection, telecommunication industry
PDF Full Text Request
Related items