Analysis of Seismic Data in Sumatra using Robust K-Means Clustering

Abstract

Indonesia is located within the Pacific Ring of Fire and frequently experiences significant seismic activities, rendering the region susceptible to hazards. Specifically, Sumatra is an island in the western part of the country, near the Eurasian and Indo-Australian tectonic plates. Over the past five years, an observable uptick in seismic events has been recorded in Sumatra. This research aimed to cluster the Sumatra region’s seismic data using the k-means algorithm and its extensions, including trimmed and robust sparse k-means, to determine the characteristics and patterns of seismic events. The k-means clustering algorithm operates effectively on many data but needs to work better in the presence of outliers. Meanwhile, the data identification reports the presence of outliers in the seismic data. The clustering analysis identified two main clusters, supported by multivariate and spatial outlier detection during preprocessing. The first cluster, encompassing 62% of seismic events, is located offshore near the Mentawai seismic gap, characterized by shallow depths (33–41 km) and magnitudes of 4.5–5.0 Ms. The second cluster, representing 28% of events, includes both mainland and offshore regions, associated with the Sumatran Fault system and slab deformation zones, at moderate depths (54–154 km) with magnitudes of 4.3–4.4 Ms. Rare deep-focus events exceeding depths of 214 km were identified as outliers. Evaluation using Silhouette, Davies-Bouldin, and Dunn indices determined that k=2 was the optimal number of clusters. This study contributes by integrating robust clustering methods to handle outliers, enhancing the reliability of seismic data analysis. This study demonstrates the value of applying trimmed and robust sparse k-means algorithms to improve clustering performance in regions with complex tectonic activity.