Irma Kurniawati, M. Aryono Adhi, Supriadi, Sulhadi, Hanifullah Hafidz Arriza, Marzuki Sinambela
Accurate identification of earthquake clusters is essential for understanding seismic behavior and improving hazard mitigation strategies in tectonically complex regions. While clustering algorithms have been widely applied to seismic data, few studies have conducted a comprehensive comparison of multiple methods in Central Java - a region with overlapping fault zones and heterogeneous seismic activity. This study addresses that gap by evaluating four clustering techniques: Density-Based Spatial Clustering of Applications with Noise (DBSCAN), Hierarchical DBSCAN (HDBSCAN), Spatio-Temporal DBSCAN (ST-DBSCAN), and Fuzzy C-Means (FCM). A dataset comprising 3,175 earthquake events from 2009 to 2023 was sourced from Indonesia's Meteorological, Climatological, and Geophysical Agency (BMKG). Standard preprocessing and grid search optimization were applied to tune parameters consistently across all models. Clustering quality was assessed using Silhouette Score, Calinski-Harabasz Index, and Davies-Bouldin Index. DBSCAN produced the most compact and well-separated clusters (Silhouette =0.82; CH= 1828.66; DB=0.28) but excluded 96.47% of events as noise. HDBSCAN achieved broader spatial coverage with minimal noise, while ST-DBSCAN effectively captured short-term temporal swarms, and FCM modeled overlapping tectonic boundaries. Although no single method proved universally optimal, DBSCAN emerged as the most precise in isolating dominant seismic features. A hybrid approach is recommended to balance spatial compactness with data inclusivity, supporting more robust seismic hazard mapping in complex regions like Central Java. © 2025 IEEE.
Universitas Negeri Semarang, Faculty of Mathematics and Natural Sciences, Semarang, Indonesia; Stmkg, Tangerang, Indonesia; Stmkg, Tangerang, Indonesia
