Dewintha Tresna Reza, Sunarno, Supriadi, Siti Wahyuni, Hanifullah Hafidz Arrizal, Marzuki Sinambela
This study explores the use of unsupervised learning algorithms - K-Means, K-Medoids, and Fuzzy C-Means (FCM) - to identify spatial patterns in 6,005 earthquake events in West Java from 2009 to 2023. Clustering was based on location, depth, and magnitude, with performance evaluated using Silhouette Score, Calinski-Harabasz Index, and Davies-Bouldin Index. K-Means produced three well-separated clusters aligned with major tectonic structures, achieving the highest Silhouette Score (0.31). K-Medoids offered finer spatial resolution and the best separation (Davies-Bouldin Index: 1.05), effectively highlighting localized seismic zones. FCM provided the highest Calinski-Harabasz Index (2227.74), capturing transitional regions through overlapping clusters. All models successfully classified the data without noise. The results show that K-Means is optimal for broad seismic zoning, while K-Medoids and FCM offer complementary insights into regional fault complexity and uncertainty. These findings support the integration of clustering methods into regional seismic hazard analysis. © 2025 IEEE.
Universitas Negeri Semarang, Faculty of Mathematics and Natural Sciences, Semarang, Indonesia; STMKG, Undergraduate Program in Applied of Instrumentation Meteorology, Climatology, and Geophysics, Tangerang, Indonesia; STMKG, Program in Applied of Instrumentation Meteorology, Climatology, Geophysics, Tangerang, Indonesia
