<![CDATA[This study proposes a hybrid Ensemble RNN–Random Forest (RNN–RF) model for short-term earthquake prediction based on spatio-temporal seismic data from the Sulawesi–Maluku region. The purpose of this research is to develop a lightweight and interpretable machine learning framework that integrates temporal and spatial features using local datasets provided by the Manado Geophysical Station of BMKG. The methodology includes six stages: data acquisition, preprocessing, feature engineering, model development, ensemble integration, and evaluation. The RNN captures sequential dependencies in seismic activity, while the Random Forest learns spatial and contextual relationships such as fault proximity and event clustering. The ensemble fuses probabilistic outputs (0.75 RNN and 0.25 RF) followed by domain-based calibration using mean magnitude, event frequency, and fault distance. Experimental results show that the proposed ensemble achieved F1 = 0.89 and AUC = 0.975, outperforming individual RNN and RF models in predictive stability and accuracy. The model demonstrates that integrating domain-specific adjustments enhances both recall and precision, while maintaining interpretability for operational deployment. This study contributes to explainable AI in seismology by bridging deep temporal modeling with geophysical reasoning, offering a scalable approach for early-warning applications in Indonesia.]]>
Copyrights © 2025
