Time-Series Forecasting of the Pazarcık Earthquake Using LSTM, Transformer and RNN Models

The Earth's internal structure and mitigating seismic hazards are very important for understanding for earthquake prediction and seismic wave analysis. In this study, we studied with different deep learning models for earthquake time series prediction using Broadband Teleseismic Data from the USGS database. This dataset consists of 1000 seismic records in SAC format with long-period seismic waves from global earthquakes. The aim of this study was to test LSTM and RNN models with LSTM Transformer to predict the next time step based on previous seismic waves. In this study, model performances was evaluated with Mean Square Error (MSE), Mean Absolute Error (MAE) and R² Score. In conclusion, the LSTM Transformer+RNN model achieves the lowest error rates and presents its effectiveness in learning both short-term dependencies and long-term correlations in seismic data. At the same time, this study can also provides to the advancement of deep learning applications in seismology and the improvement of the prediction capabilities of earthquake monitoring systems.