An empirical comparison of techniques for the class imbalance problem in churn prediction

Class imbalance brings significant challenges to customer churn prediction. Many solutions have been developed to address this issue. In this paper, we comprehensively compare the performance of state-of-the-art techniques to deal with class imbalance in the context of churn prediction. A recently developed expected maximum profit criterion is used as one of the main performance measures to offer more insights from the perspective of cost-benefit. The experimental results show that the applied evaluation metric has a great impact on the performance of techniques. An in-depth exploration of reaction patterns to different measures is conducted by intra-family comparison within each solution group and global comparison among the representative techniques from different groups. The results also indicate there is much space to improve solutions' performance in terms of profit-based measure. Our study offers valuable insights for academics and professionals and it also provides a baseline to develop new methods for dealing with class imbalance in churn prediction. (C) 2017 Elsevier Inc. All rights reserved.