A Behavioral-Enhanced RFM Framework for Customer Segmentation and Churn Prediction

Abstract

Customer segmentation is a fundamental task in customer relationship management and marketing analytics.The traditional RFM (Recency, Frequency, Monetary) model is widely used; however, it does not fully capture complexbehavioral patterns of customers.This paper proposes a Behavioral-Enhanced RFM (BRFM) framework that incorporatesadditional customer interaction features, including average quantity per order, product diversity, and purchase regularity.The framework applies feature engineering and normalization techniques, followed by K-Means clustering for customersegmentation. Clustering performance is evaluated using the Elbow method and Silhouette score.Furthermore, a predictivemodeling approach is employed using machine learning algorithms to estimate customer churn. Experimental resultson the Online Retail dataset indicate that, although the inclusion of behavioral features does not significantly improveclustering compactness compared to traditional RFM, it substantially enhances predictive performance. The BRFM-basedmodel achieves higher precision, F1-score, and ROC-AUC in churn prediction.These findings demonstrate that integratingbehavioral features provides a more comprehensive understanding of customer behavior and supports more effectivedecision-making for targeted marketing and customer retention strategies.