On the use of VMD-LSTM neural network for approximate earthquake prediction

Earthquake prediction has been widely studied in many fields using various technologies, including machine learning, which is able to explore the underlying information of data. This study adopts machine learning for earthquake prediction, and employs the well-studied long short-term memory (LSTM) neural networks to predict the earthquake occurrence time, longitude, latitude, and magnitude. A variational mode decomposition (VMD) approach is also used to improve the precision of the precision. The proposed model is referred to as VMD-LSTM. The used datasets consist of earthquake catalogs from 1935 to 2023 in the Taiwan region, where earthquakes having magnitudes greater than 5.0 are studied. Four VMD-LSTMs are constructed to predict the earthquake occurrence time, longitude, latitude, and magnitude. The experimental results show that the proposed model has high performance on the test set. The results are also compared with those obtained by the LSTM model without VMD, verifying that the VMD-LSTM model leads to higher performance. This study introduces a novel model for the prediction of earthquake magnitude, location, and time. It also demonstrates that the VMD method can be combined with various neural networks to increase the prediction accuracy.