An application of rough set theory to predict telecom customer churn

Abstract

The current paper applies algorithms of machine learning to predict customer churn. The study employs 211,777 instances in the telecommunication sector with six attributes employed, e.g., data, length of stay, top-up, external communication, handset of phone, and churn. Although the rules generation of Naïve Bayes, J48 (Decision Tree), and Decision Table are used, the algorithm of Decision Table is the best candidate to support churn prediction due to its highest accuracy rate of 88.8%. The finding also confirms the role of the external communication of subscribers through calls and messages (in two ways) by other subscribers from the different telecom operators influencing the subscriber's churn. The finding is a significant contribution to the telecom operators to predict churn. In particular, it comes at a time when government regulations have been adjusted to allow phone users to change networks from different service providers, but keep the same phone number.)