Clustering Analysis of Customer Behavior Using Data-Driven Methods

Abstract

Customer behavioral data can be used to build user and market profiles for targeted marketing. However, heterogeneity across customers makes these data difficult to organize into meaningful groups. This paper investigates a data-driven approach to customer categorization based on RFM characteristics (recency, frequency, and monetary value). Three representative unsupervised learning algorithms, including K-means, density-based spatial clustering of applications with noise (DBSCAN), and Ward hierarchical clustering, are examined in terms of their objective functions, distance assumptions, parameter sensitivity, and validation indices. The workflow covers feature transformation and standardization, cluster evaluation, and business interpretation using 1,000 customer behavior records. K-means produces four balanced and interpretable groups whose structure is broadly consistent with hierarchical clustering, although its internal validity is slightly lower. DBSCAN separates the retained core observations well, but it labels a relatively large number of customers as noise. Therefore, a high silhouette score does not necessarily imply broad business applicability. The results indicate that algorithm selection should jointly consider mathematical fit, coverage, robustness, and marketing actionability rather than rely on a single index.