Analisis Perbandingan Kinerja Algoritma K-Means dan K-Medoids dengan Reduksi Dimensi PCA pada Indikator Kesehatan dan Sosial

Abstract

Public health in West Java faces complex challenges, including disparities in healthcare access, malnutrition, and socio-economic inequalities across districts. These conditions require data-driven analysis to identify patterns of disparity and provide evidence-based guidance for policy intervention. This study aims to cluster districts/cities in West Java based on health and social indicators using Principal Component Analysis (PCA) for dimensionality reduction, followed by K-Means and K-Medoids algorithms for clustering. Data from 27 districts/cities during 2019–2024 were analyzed after standardization. PCA extracted two principal components explaining 61.4% of the total variance. Scree plot and silhouette results indicated three optimal clusters. Comparative analysis revealed that the average silhouette score of K-Means was 0.31, while K-Medoids achieved a higher score of 0.34, suggesting more stable and robust partitioning against outliers. In 2024, Cluster 1 consisted of regions with adequate healthcare facilities and lower prevalence of underweight children; Cluster 2 grouped regions with limited health infrastructure and higher malnutrition problems, while Cluster 3 showed intermediate conditions. Therefore, K-Medoids outperformed K-Means by producing more consistent clustering across years. These findings offer practical recommendations: Cluster 2 should be prioritized for interventions such as improving primary healthcare access and nutrition programs, Cluster 1 requires maintenance of service quality, and Cluster 3 should be targeted for gradual reinforcement.