Earthquake prediction using a hybrid deep learning model
نویسندگان
1 MSc. Student, Department of Computer Engineering, University of Meybod, Meybod, Iran
2 Assistant Prof, Department of Computer Engineering, University of Meybod, Meybod, Iran
3 Assistant Prof, Department of Computer Engineering, University of Meybod, Meybod,
4 Associate Prof, Department of Computer Engineering, University of Meybod, Meybod, Iran
doi
10.22091/jemsc.2025.13249.1282چکیده
In This study applies deep learning methods to predict earthquakes with magnitudes over 5.5 using a dataset of over 23,000 seismic events recorded from 1990 to 2024 in the Sarpol-e Zahab region. Several models were developed, including CNN, LSTM, Transformer, and a hybrid model combining CNN, LSTM, and Attention layers. The hybrid model demonstrated superior performance by capturing spatial patterns, temporal dependencies, and attention-based context, achieving 99.34% accuracy and a 0.0285 loss on the test set. Final evaluation yielded 99.51% accuracy, 96.59% precision, 93.92% recall, and a 95.24% F1-score, highlighting the model’s effectiveness in predicting potential earthquakes within a 30-day window., the results indicate that hybrid deep learning models offer valuable tools for developing intelligent early warning systems. this research contributes to improving seismic preparedness and risk reduction strategies in earthquake-prone regions.
