The e-commerce membership platforms that provide membership services through subscription- based models are currently experiencing a serious situation of customer churn. This paper suggests the use of a Counterfactual E-Commerce Churn Prediction (CECP) model, which will combine the strengths of machine learning, explainable AI, and causal inference techniques to predict and prevent customer churn. The model will employ XGBoost for predicting customer churn, SHAP for explaining the model, and DiCE for counterfactual explanation. In addition, the T-Learner uplift modeling approach will be employed to determine the individual treatment effects and customers who can benefit from customer retention strategies in the most optimal way. The experiments performed on a large e-commerce membership dataset suggest that the proposed approach will yield an accuracy of 89% and area under the curve of 0.93, and the simulated intervention strategies will actually decrease customer churn by 23% using simulated intervention strategies. The findings suggest the efficacy of using predictive models in conjunction with a counterfactual reasoning model to develop data-driven and personalized customer retention strategies.
S et al. (Thu,) studied this question.