Machine learning ensemble techniques for customer lifetime value prediction in e-commerce businesses: A review

Online retailers increasingly depend on precise Customer Lifetime Value (CLTV) estimation for data-driven Customer Relationship Management. Traditional frameworks, including RFM heuristics, probabilistic Buy-Till-You-Die models, and Markov chain architectures, offer interpretability but fail when confronted with highly skewed, zero-inflated behavioral datasets common in digital commerce. This review synthesizes the transition from legacy methods to ensemble learning techniques, focusing on homogeneous bagging (Random Forests), sequential boosting (GBM, XGBoost, CatBoost), and heterogeneous stacked generalization. The study identifies critical success factors, performance trends across asymmetric optimization targets, and dataset variations. This study serves as a definitive reference to guide future CLTV deployment pipelines and validation rigor.