Customer Segmentation Using Clustering Algorithms: A Comparative Evaluation of K-Means, Hierarchical, and DBSCAN Methods

Abstract

Customer segmentation, the grouping of customers with similarcharacteristics, is a cornerstone of targeted marketing and customerrelationship management. Clustering, an unsupervised data-mining technique, provides a natural means of discovering such segments from customer data. This paper presents a comparative evaluation of three widely used clustering algorithms, namely K-Means, agglomerative hierarchical clustering, and DBSCAN, applied to customer data for segmentation. The algorithms were compared using internal validity measures, including the silhouette score and the Davies-Bouldin index, and the resulting segments were profiled and interpreted. The results show that K-Means produced compact and well-separated clusters and was the most efficient, hierarchical clustering offered an interpretable dendrogram but scaled poorly, and DBSCAN identified clusters of arbitrary shape and isolated outliers but was sensitive to its parameters. K-Means achieved the best overall validity for this dataset, with an optimal number of segments determined from the silhouette analysis. The study clarifies the strengths and limitations of each algorithm and offers guidance on selecting a clustering method for customer segmentation. KEYWORDS: Customer Segmentation, Clustering, K-Means, HierarchicalClustering, DBSCAN, Silhouette Score, Unsupervised Learning