Spatial Clustering of Seismic Events in Bali and West Nusa Tenggara

Abstract

The precise delineation of seismic source zones is fundamental to seismic hazard assessment, particularly in complex tectonic settings like around Bali and West Nusa Tenggara (NTB). Traditional zoning methods often struggle to capture the irregular geometries of active faults, leading to potential inaccuracies in hazard estimation. This study presents a comparative analysis of partition-based (K-Means) and density-based (DBSCAN) clustering algorithms to identify active fault structures using earthquake data recorded from 2015 to 2025. The dataset, comprising standardized hypocentral coordinates, was processed to evaluate the ability of each algorithm to segregate active ruptures from background seismicity. Results indicate that K-Means (k=3) imposed artificial, spherical boundaries that fragmented continuous geological features and misclassified distant outliers. Conversely, DBSCAN, optimized with an Epsilon (eps) of 0.25 and Minimum Points (MinPts) of 5, successfully delineated the linear continuity of the Flores Back-Arc Thrust and isolated distinct volcanic earthquake swarms while effectively filtering noise. The study concludes that DBSCAN offers superior performance for seismotectonic mapping in the Bali-NTB region, providing a more geologically realistic framework for defining seismic hazard zones than traditional partitioning methods.