<![CDATA[Disparities in regional health status remain a major challenge for data-driven public health planning in Indonesia. This study aims to develop a machine learning-based clustering approach to group 27 regencies/cities in West Java Province according to the similarity of their health and socio-demographic indicator profiles using open government data. The research follows the CRISP-DM framework, from business understanding to deployment. Five of nine indicators were converted into population-based and live-birth-based rates to remove the confounding effect of regional size differences, then selected using variance thresholding and Spearman correlation, resulting in eight final features. Five clustering algorithms were compared: K-Means, K-Medoids (PAM), DBSCAN, Agglomerative Hierarchical Clustering, and Fuzzy C-Means (FCM), with the optimal number of clusters determined through the Elbow Method and Silhouette Analysis. The results indicate k=6 as the optimal number of clusters, with K-Means, K-Medoids, and Agglomerative Hierarchical Clustering producing identical and best cluster quality (Silhouette Index 0.289; Davies-Bouldin Index 0.978; Calinski-Harabasz Index 9.52), outperforming Fuzzy C-Means (0.184), while DBSCAN failed to form meaningful clusters due to the relatively homogeneous density distribution of the data. Cluster profiles revealed significant disparities, particularly in doctor ratio, proper sanitation, and access to proper water sources: the Bekasi-Depok cluster showed the most favorable profile, while the small-city cluster (Sukabumi, Tasikmalaya, Banjar, Cirebon) showed the highest infant and maternal mortality rates despite relatively adequate health infrastructure. This study provides a methodological contribution as a foundation for developing an open-data-based regional health and socio-demographic indicator mapping system to support more targeted regional health policy.]]>
