<![CDATA[Accurate ground motion prediction is critical for seismic hazard mitigation in Indonesia, a region characterized by complex tectonic settings. Conventional Ground Motion Prediction Equations (GMPEs) often struggle to capture the non-linear attenuation characteristics of local seismicity. This study develops national-scale Machine Learning (ML) models—specifically XGBoost and Random Forest (RF)—to predict Peak Ground Acceleration (PGA) and benchmarks them against the widely used Zhao et al. [23] model. The dataset comprises 20,287 strong-motion records from 1,573 events (M 1.6–7.9) recorded by 667 stations across Indonesia between January 2023 and April 2025. Performance evaluation reveals that the XGBoost model outperforms both RF and the conventional GMPE, achieving the lowest Mean Squared Error (MSE) of 0.8885. In contrast, the conventional model showed less reliability with a significantly higher MSE of 1.5905. Feature importance analysis indicates that hypocentral distance and magnitude are the dominant predictors, while site condition (Vs30) plays a secondary role. Validation on independent test sets and recent significant seismic events confirms the model's robust generalization capability compared to conventional approaches. These findings demonstrate that ML-based approaches provide a more reliable alternative for estimating ground motion in Indonesia, offering significant potential for improving early warning systems and hazard maps.]]>
Copyrights © 2026
