Integrated seismic modelling for NPP risk assessment in Serpong: using hybrid machine learning clustering and support vector machine methods

Abstract The pursuit of clean and sustainable energy in Indonesia has intensified focus on nuclear power, positioning Serpong, West Java, as a prospective site. While previous studies have addressed regional seismicity, a comprehensive, high-resolution seismic hazard assessment – critical for the safety of nuclear power plants (NPPs) – remains underdeveloped. This study bridges the gap by implementing a hybrid machine learning framework that integrates DBSCAN, Hierarchical Clustering, and Support Vector Machine (SVM) to delineate seismic source zones and evaluate fault activity. Our methodology leverages an updated earthquake catalog up to 2024 and incorporates newly identified seismic sources to enhance clustering precision. The analysis delineates 12 distinct seismic zones with heterogeneous magnitude and depth distributions, pinpointing specific high-risk areas. Gutenberg-Richter analysis reveals anomalously low b-values within these zones, signaling a heightened potential for large-magnitude seismic events. Furthermore, the SVM model successfully captures complex, non-linear seismic patterns, significantly improving hazard prediction accuracy. The findings provide critical, data-driven insights for informed site selection, resilient structural design, and proactive risk mitigation strategies. The proposed framework establishes a robust and transferable paradigm for preliminary seismic hazard evaluation of critical infrastructure in seismically active regions globally.