Application of decision trees (DTs) method for magnitude prediction of earthquakes in Türkiye

The applicability of decision tree methods for predicting the magnitude of future earthquakes in Turkey has been investigated. Given the country's high seismic risk, latitude, longitude, and depth data were utilized to estimate earthquake magnitudes. This study employed earthquake records from Boğaziçi University Kandilli Observatory, covering the period from 1900 to 2024. A prediction model was developed using the C4.5 decision tree algorithm. The dataset, consisting of earthquakes ranging from magnitude 3.0 to 8.0, was divided into 80% training and 20% testing data. The model demonstrated a generalization capability with an accuracy of 88%; however, it struggled to predict earthquakes with magnitudes of 6.0 and above due to class imbalance. While predictions for small and moderate earthquakes were more accurate, performance declined for large-magnitude events. The relationship between depth and magnitude was examined using the Chi-Square test (χ²=5674.81, p=0.0), confirming a statistically significant correlation. Shallow-depth earthquakes (0-30 km) were found to be more frequent in lower magnitudes, whereas 92% of large earthquakes occurred within the same depth range. Geospatial distribution analyses indicated that the model's predictions aligned with seismically active zones such as the North Anatolian Fault and the Aegean Graben System. Although decision trees provide a transparent and interpretable approach, limitations in predicting rare large earthquakes remain a challenge. Future research should integrate deep learning techniques, incorporate geological and tectonic parameters, and establish regional prediction centers to enhance the accuracy and reliability of earthquake forecasting.