Learning trends in customer churn with rule-based and kernel methods

Abstract

In the present article an attempt has been made to predict the occurrences of customers leaving or 'churning' a business enterprise and explain the possible causes for the customer churning. Three different algorithms are used to predict churn, viz. decision tree, support vector machine and rough set theory. While two are rule-based learning methods which lead to more interpretable results that might help the marketing division to retain or hasten cross-sell of customers, one of them is a kernel-based classification that separates the customers on a feature hyperplane. The nature of predictions and rules obtained from them are able to provide a choice between a more focused or more extensive program the company may wish to implement as part of its customer retention program.