Earthquakes magnitude prediction using deep learning for the Horn of Africa

Abstract

Earthquakes are vibrations of the Earth's surface that can cause ground shaking, fires, tsunamis, landslides and fissures. These natural phenomena can cause destruction and kill lives. When there is a possibility of an earthquake, an accurate prediction can save lives and avoid infrastructure damage. Due to the probabilistic nature of an earthquake occurring and the challenge of achieving an efficient and dependable model for an earthquake prediction, efforts to predict earthquakes have been met with mixed results. Therefore, new methods are constantly sought. A deep learning-based technique, specifically a transformer algorithm, was applied to predict earthquake magnitudes using available data for the Horn of Africa. The problem was formulated as multi-variant time series regression, and predictions were made for earthquakes magnitudes greater than or equal to 3 for the next three months. A comparison of the results was made with the output obtained from long short-term memory (LSTM), bidirectional long short-term memory (BILSTM), and bidirectional long short-term memory with attention (BILSTM-AT) models. The results showed that the transformer model outperformed the other three models with 0.276, 0.147, 0.383, 28.868% mean absolute error (MAE), mean squared error (MSE), root mean squared error (RMSE) and mean absolute percentage error (MAPE) respectively in predicting earthquake magnitudes in the Horn of Africa.