Customer Segmentation using WRFM-Based Feature Representation with K-Means and PCA

Abstract

This study investigates customer segmentation using a Weighted Recency, Frequency, and Monetary (WRFM)-based feature representation combined with Principal Component Analysis (PCA) and K-Means clustering. A dataset consisting of 286 transactions from 100 customers was transformed into RFM variables and normalized using Min–Max scaling to ensure comparability. Entropy-based weighting was then applied to construct WRFM features, providing a balanced representation of customer behavior. PCA was employed to reduce dimensionality and eliminate redundant information prior to clustering. Four clustering methods—K-Means, PCA-enhanced K-Means, Gaussian Mixture Model (GMM), and Hierarchical Clustering—were evaluated using Silhouette Score, Calinski–Harabasz Index, and Davies–Bouldin Index. The results indicate that PCA-enhanced K-Means achieved the best performance (Silhouette = 0.473, Calinski–Harabasz = 144.817, Davies–Bouldin = 0.725), demonstrating improved cluster compactness and separation. The findings show that combining WRFM-based features with PCA effectively reduces bias and improves clustering quality. The segmentation results identify four distinct customer groups, including high-value loyal customers, potential customers, moderate customers, and at-risk customers. These insights support targeted marketing strategies such as retention programs, personalized promotions, and customer relationship management. This study does not propose a new clustering model but focuses on empirical evaluation and applied validation of existing techniques. The approach is scalable and applicable to other domains involving multidimensional customer data.