Investigating Decision Tree in Churn Prediction with Class Imbalance

Bing Zhu, Guicai Xie, Yuan Yuan, Yiqin Duan

Abstract: Class imbalance presents significant challenges to customer churn prediction. Traditional machine learning algorithms like decision tree tend to be biased towards majority class. In this paper, we comprehensively study the performance of decision tree in churn prediction with class imbalance. We investigate the issue of pruning setting and optimal sampling strategy based on a recently developed expected maximum profit criterion. The experiments present some different conclusions from the previous research when the area under the ROC curve is used and the optimal sampling strategy are recommended. Our findings provides a useful guideline for usage of decision tree in churn prediction.