<![CDATA[Poverty in North Sumatra exhibits considerable spatial heterogeneity, yet comparative evidence on the performance of clustering algorithms for identifying multidimensional regional poverty patterns remains limited. This study aimed to compare K-Means and K-Median in clustering 33 regencies and municipalities using five poverty-related indicators averaged over 2022–2025: percentage of poor population, poverty line, poverty gap index, poverty severity index, and adjusted per capita expenditure. A quantitative descriptive-comparative design was applied using secondary data from the Central Statistics Agency of North Sumatra Province. All variables were transformed through Min–Max normalization, the Elbow Method was used to determine the optimal number of clusters, and clustering performance was evaluated using the Silhouette Coefficient. The analysis identified three clusters for both algorithms. K-Means produced cluster sizes of 28, 2, and 3 regions, whereas K-Median generated 21, 2, and 10 regions. A notable finding was that both methods consistently placed South Nias and North Nias in the same distinct cluster, characterized by comparatively high poverty gap and poverty severity values and low adjusted per capita expenditure. K-Means achieved a higher Silhouette Coefficient than K-Median, with values of 0.2953 and 0.1515, respectively, indicating comparatively better overall clustering performance. These findings show that algorithm selection influences both cluster structure and the substantive interpretation of multidimensional poverty patterns. The study provides empirical support for targeted and context-sensitive regional poverty alleviation policies across the diverse administrative areas of North Sumatra.]]>
