Churn prediction in subscriber management for mobile and wireless communications services

Abstract

Subscriber churn is a concern of customer care management for most of the mobile and wireless service providers and operators due to its associated costs. This paper explains our work on subscriber churn analysis and prediction for such services. We work on data mining techniques to accurately and efficiently predict subscribers who will change-and-turn (churn) to another provider for the same or similar service. The dataset we use is a public and real dataset compiled by Orange Telecom for the KDD 2009 Competition. Number of teams achieved high scores on this dataset requiring a significant amount of computing resources. We are aiming to find alternative methods that can match or improve the recorded high scores with more efficient and practical use of resources. In this study, we focus on ensemble of meta-classifiers which have been studied individually and chosen according to their performances.