Machine Learning (ML) is a branch of artificial intelligence that leverages historical data to learn and make accurate predictions about future outcomes. This mathematical capability has been applied to predict earthquakes in the Iranian Plateau. This research aims to investigate the performance of these techniques and their differences according to the variations in seismic behavior in the seismotectonic province of the Iranian Plateau. Nine seismic indices have been selected based on the physics of rupture. Three models—Artificial Neural Networks (ANN), Random Forest (RF), and Support Vector Machines (SVM)—have been employed, and earthquakes with magnitudes of 5 to 6 occurring over one month have been formulated as a binary classification task. Four statistical measures (accuracy, precision, FI- score, and recal) have been estimated to evaluate the optimal technique. The results indicate that the Zagros seismotectonic province contains the most extensive dataset, enhancing the accuracy of all methods applied. While the Artificial Neural Network (ANN) method outperforms others at a threshold magnitude across Alborz-Azerbaijan, Kopeh-Dagh, Makran, and Zagros provinces, the Random Forest (RF) method exhibits superior performance in the Central-East Iran province. At the same time, the highest value is obtained for all seismotectonic provinces using the RF method. However, the SVM method also serves as a good alternative to the RF method in predicting rare events in limited seismic catalogs of the Makran and Kopeh-Dagh seismotectonic provinces. Increasing knowledge about the efficiency of these methods is a vital step toward developing applicable forecasting methods in active seismic areas and improving crisis management services.
Salma Ommi (Sun,) studied this question.