Learning Earthquake Patterns over Mediterranean Grids Using Random Forests and Gradient Boosting

Abstract

This study applies Random Forest (RF) and Gradient Boosting (GBoost) methods to predict earthquake magnitudes across Mediterranean zones defined by a $10^{\circ} \times 10^{\circ}$ spatial grid. Model performance is assessed using Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE). Both models achieve their best results in the 40–50° latitude and 0°10° longitude zone, where the Random Forest model reaches an MSE of 0.276, an RMSE of 0.526, and an MAE of 0.388, while Gradient Boosting performs slightly better with an MSE of 0.244, an RMSE of 0.494, and an MAE of 0.357. These results highlight the strong predictability and high-quality seismic data in this region, demonstrating that tree-based ensemble methods effectively model regional seismic patterns and contribute to improving earthquake forecasting across the Mediterranean region.