Prediction of Peak Ground Acceleration for The Bengkulu Region Using Artificial Neural Network

Febriani, Yeza
Fatkhurrochman, Fatkhurrochman
Yunita, Farida
Farid, Muhammad
Apriniyadi, Mohammad
Saleh, Arif Rahman

JURNAL ILMIAH PENDIDIKAN FISIKA AL BIRUNI
Vol 14 No 1 (2025): Jurnal Ilmiah Pendidikan Fisika Al-Biruni

This study aimed to predict Peak Ground Acceleration (PGA) values for the Bengkulu region using an Artificial Neural Network (ANN) model. The ANN model utilized earthquake parameter inputs, including magnitude, depth, and hypocenter distance, with soil PGA data collected from the Bengkulu City area. The PGA values were estimated using a neural network model, with the results optimized, validated, and evaluated for performance. The model accurately predicted PGA for large-magnitude earthquakes (R² = 0.99 for magnitudes 7.9–6.5). However, its performance declined significantly for smaller magnitudes (R² = 0.0141 for magnitude 4), reflecting challenges in accurately capturing input parameters, like focal depth and epicentre distance for low-magnitude events. Across a magnitude range of 4.0 to 7.9, the model achieved an overall R² value of 0.99, indicating high accuracy, particularly for larger magnitudes. However, the model's performance declined for lower magnitudes, with R² values dropping significantly, attributed to inaccuracies in input parameters, such as focal depth, epicentre distance, and period. The study provided logarithmic equations for each magnitude range tailored to the seismic characteristics of Bengkulu City, highlighting the importance of localized PGA prediction models. The findings suggest the potential effectiveness of the ANN model for improving earthquake early warning systems and seismic risk management in Bengkulu City under simulated conditions, particularly for large-magnitude earthquakes (R² = 0.99). However, the model’s limitations in predicting PGA for low-magnitude events (R² = 0.0141) highlight the need for further refinement before real-world implementation. This research contributes to the growing body of knowledge by validating and refining the ANN approach for region-specific seismic conditions, offering a practical tool for local authorities and disaster management agencies. Future research should improve the model's accuracy for low-magnitude earthquakes and explore hybrid machine learning techniques to enhance predictive.