Shopping pattern segmentation: HAC versus K-Means performance analysis

Abstract

Despite widespread use in consumer analytics, clustering techniques remain underutilized for analyzing household basic food commodity consumption patterns, particularly for developing localized retail strategies and targeted food security policies in resource-constrained contexts. This study addresses this practical gap by systematically comparing Hierarchical Agglomerative Clustering (HAC) and K-Means performance on essential consumption patterns across seven commodities: bread, vegetables, fruit, meat, poultry, milk, and wine. Using dual validation metrics, Silhouette Coefficient and Davies-Bouldin Index, we evaluate clustering effectiveness specifically for small-scale household datasets typical of regional food policy environments. HAC demonstrated superior cluster stability (Silhouette score = 0.2936, DBI = 0.8977) compared to K-Means (0.2912, 0.9871), enabling identification of three actionable consumption segments, namely budget-conscious households with economical protein consumption, high spender households with premium patterns across categories, and balanced/selective households preferring bread and wine. These empirically-derived segments provide implementable frameworks for food subsidy targeting, inventory optimization in local retail contexts, and nutrition intervention program design. The findings demonstrate that methodologically rigorous clustering analysis yields policy-relevant household segmentation even with constrained data, offering practical guidance for evidence-based food security interventions where basic commodity consumption directly informs resource allocation decisions.