Research And Application Of Data Mining In Telecom Churn Analysis

With the reorganization and reform of the telecommunications industry, the telecom operators start operating a wide range of telecommunication services. At present, China Telecom have set up all forms of business combination, China Unicom also have a similar combination of business. Facing with a number of telecommunications enterprises in a variety of business, consumers will evaluate and select the most suitable one. The use of combination of business not a single business, customers would produce a higher ARPU (Average Revenue Per User) value, the behavior of consumers' churning can cause the greater losses. Now, the telecommunications industries are being integrated, combination of business is an inevitable trend, analysis of telecom customer churning is also the inevitable work in future.Telecom Customer Churn System is a secondary analysis system of customer relationship management; it studies consumers' communication behavior from the history data and models the potential rule. Business analysts classify the existing customer by using model rules, identify possible churning customers from normal customers and implement customer care operations, in order to extend the service of customers and ensure that high-value customers continue to use selected business. Model rules can be used for the development of new business and the adjustment of existing business, and then telecommunications companies have the higher rate of on-line customer and strengthen the competitiveness.The main contents of this thesis are data mining research and application of model about customers' churning, especially improved cascade algorithm based on unbalanced data sets. The thesis has done the work of the following aspects:1) Development and improvement of the classification algorithms used in the current application.2) Strategies of dealing with unbalanced data sets and Classification algorithms being used in telecom customers churn analysis system.3) Propose the revised cascade decision-tree algorithm based on unbalanced data sets and verify the performance of improved algorithm. 4) Revise the system of telecom customer churning analysis using the new algorithmand Test the trial version.Above all, author propose the revised cascade decision-tree algorithm based on unbalanced data set, revise the system of telecom customer churn analysis using the new algorithm and improve the performance of the original system. The thesis provides the reference for future research.