<![CDATA[Non-communicable diseases (NCDs) account for approximately 74% of all global deaths annually and impose a disproportionate burden on developing nations, including Indonesia. In Banten Province, epidemiological surveillance data reveal highly heterogeneous distributions of NCD behavioral risk factors across 149 sub-districts (aggregated from facility-level surveillance records), complicating the design of targeted public health interventions. This study proposes and evaluates four metaheuristic-enhanced Hierarchical Agglomerative Clustering (HAC) frameworks—GA-HAC, PSO-HAC, GWO-HAC, and WOA-HAC—applied to five behavioral risk indicators: excessive fat/oil consumption (efc), excessive salt consumption (esc), excessive sugar consumption (esuc), tobacco smoke exposure (etcs), and active smoking (smoke). Each metaheuristic optimizes a composite fitness function combining the Silhouette Score and the Davies-Bouldin Index on log1p-transformed features. Across 20 independent runs, WOA-HAC achieves the best mean Silhouette Score (0.4745) and lowest mean Davies-Bouldin Index (0.6795), significantly outperforming PSO-HAC and GWO-HAC on both criteria (Kruskal-Wallis, p<0.001), though no single algorithm dominates the Calinski-Harabasz criterion. The resulting three-tier solution reveals a Low Risk tier (n=25, 16.8%), a smoking-predominant tier concentrated in Kabupaten Pandeglang (n=26, 17.4%), and a Very High Risk tier spanning all eight administrative units (n=98, 65.8%) with majority-level prevalence across all five indicators simultaneously. Elevated NCD risk in Banten Province thus appears widespread rather than confined to a small urban enclave, warranting province-wide, multi-component prevention programming.]]>
