Research Review on Customer Life Cycle Value Prediction Based on RFM Analysis and Ensemble Learning

Abstract

Accurate prediction of customer lifetime value (CLV) is a fundamental point relied upon for optimizing resources and formulating marketing plans. The traditional RFM method can illustrate the general characteristics of customer behavior, but it is difficult to handle nonlinear problems; a single ensemble learning method, although highly accurate, lacks interpretability. Therefore, this paper combines RFM and ensemble learning to examine their integrated form in terms of CLV, discussing the integration methods of input features, customer classes, and Stacking separately, and concentrating on analyzing the applicable scenarios and actual effects of random forest, LightGBM, and Stacking models. This also points out that existing research still has common deficiencies in RFM time window adaptation, customer value evolution, and model transparency, and proposes the possibility of improving from dynamic features, multi-model cooperation, and interpretability perspectives, providing a reference for subsequent research.