Abstract:

Data mining is a popular technology of finding regularity through analysis a large amount of data, while the customer churn analysis system is a new application to realize customer churn model based on data mining technology. Bayesian networks have good logic, predictability, and a great advantage to solve the problem of uncertainty and incompleteness as well as dealing with complex issues. This paper adopts Bayesian networks for customer churn analysis, mining the result in churn characters, in order to help decisionmaking manager formulate corresponding detainment strategy. In order to compare the performance of Bayesian network, this paper constructed two improved Bayesian network model, that is, the tree augmented naive Bayesian network, Markov Blanket Bayesian network model, and compare the classification performance with the neural network model. Experimental results show that classification performance of Markov blanket Bayesian network model improved to some extent.

Key words: Bayesian network;feature selection;customer churn