Predicting the magnitude of an impending earthquake using deep learning techniques

Earthquakes are one of nature's most devastating disasters. Earthquake prediction is critical in seismology since its success can save lives, property, and infrastructure. Numerous technologies have been proposed to address this issue, including mathematical analysis, artificial intelligence, and machine learning algorithms. Unfortunately, due to earthquakes' dynamic and spontaneous nature, they frequently fail to provide positive results. The study uses deep learning techniques to predict the magnitude of an impending earthquake using eight mathematically calculated seismic indicators derived from Japan, Indonesia, and the Hindu-Kush Karakoram Himalayan (HKKH) region's earthquake catalogs. Three deep learning techniques, including Long Short-Term Memory (LSTM), Bi-directional Long Short-Term Memory (Bi-LSTM), and self-attention-based transformer, have been implemented to model the associations between calculated seismic indicators and potential earthquake incidents. These models have been evaluated with well-known matrices such as the Mean Absolute Error (MAE), Mean Squared Error (MSE), log-cosh loss, and Mean Squared Logarithmic Error (MSLE). The value of these cost functions converges to a small number for all models, indicating that these models effectively predict earthquake magnitudes. When these models were fed with an unknown test dataset from Japan, the LSTM model performed best with the least deviation metrics (MAE = 0.060, MSE = 0.006, log cosh = 0.042 and MSLE = 0.003). Similarly, the Bi-LSTM model delivered the ideal result for the Indonesia earthquake catalog (MAE = 0.073, MSE = 0.009, log cosh = 0.016, and MSLE = 0.009), while the transformer model produced the optimal result for the HKKH region (MAE = 0.062, MSE = 0.006, log cosh = 0.043, and MSLE = 0.003). Predicting earthquake magnitude at various locations using these methodologies produces significant and positive results for magnitudes ranging from 3.5 M to 6.0 M, paving the way for the ultimate robust prediction mechanism, that has not yet been developed.