Abstract
Given that the mobile phone market in Japan is mature, it is a constant struggle for mobile phone carriers to prevent loss of their customers to other carriers, a process known as churning. To offer their customers incentives, mobile phone carriers distinguish between profitable customers and unprofitable customers and give preferential treatment, such as large discounts, to profitable customers. However, as of yet, mobile phone carriers have not been able to stop churning completely. Mobile phone carriers can use various types of transaction data such as web access history data and search words. These data are thought to reflect the preferences of customers. However, it is also a fact that carriers still cannot use these data for their own management strategies because of restrictions in the Private Information Protection Law and the Telecommunications Industry Law. In this study, we focus on web access history data and propose a method of effectively finding relationships between websites visited and churning. We also try to improve the accuracy of the prediction model for churning using web access history data.
