Predicting the Earthquake Magnitude along Zagros Fault Using Time Series and Ensemble Model

Abstract

Predicting the earthquake magnitude is a complex problem, which has been carried out in recent years. The machine learning, geophysical, and regression methods were used to predict earthquake magnitude in literature. Iran is located in a highly seismically active area; thus, earthquake prediction is considered as a great demand there. In this study, two time series algorithms are utilized to predict the magnitude of the earthquake based on previous earthquakes. These models are autoregressive conditional heteroscedasticity (GARCH), autoregressive integrated moving average (ARIMA), and the combination of ARIMA and GARCH by multiple linear regression (MLR) technique (model 3). The 9017 events are used to train and predict earthquake magnitude. On the other hand, 6188 events are applied for training models, and then 2829 events are utilized for testing it. The statistical parameters, such as correlation coefficient, root mean square error (RMSE), normalized square error (NMSE), and fractional bias, are calculated to evaluate the accuracy of each model. The results demonstrate that the ARIMA and model 3 can predict future earthquake magnitude better than other models.