Counterfactual learning in customer churn prediction under class imbalance

作者：Y. Li, Xue Song, T.Y.C. Wei, Bing Zhu · 年份：2023 · DOI：10.1145/3627377.3627392 · 被引用次数：6 · 研究领域：Customer churn and segmentation、Imbalanced Data Classification Techniques、Customer Service Quality and Loyalty

Nowadays, in churn prediction, many decision-makers are trying to obtain knowledge from models through interpretation techniques due to the non-transparency of black-box. In these techniques, the counterfactual explanation generated by the counterfactual learning method is an easy-to-understand and quantitative explanation of a single instance in black-box model prediction. Counterfactual learning relies on the transition of instances across the decision boundary, while the impact of class imbalance issue and instance position in customer-related data is insufficiently considered in recent studies. In this case, this research innovatively proposes that when generating counterfactual explanations, the impact of class imbalance issue and the instance location in customer data needs to be considered. And through comparative experiments we prove that there are obvious differences in the success rate of finding a counterfactual explanation, the distance between the counterfactual explanation and the original instance (i.e. proximity), the proportion of feature change (i.e. sparsity), and the degree of proximity support (i.e. credibility) with the original instance in different instance locations and unbalanced scenarios. In addition, in our experiments, the impact of class imbalance and instance positions vary among counterfactual methods. This research provides a reference for the application of counterfactual learning in the field of customer churn prediction and elaborates that...