A novel approach for earthquake spatial probability modeling by integration of geodetic strain into explainable artificial intelligence (XAI)

Abstract This study proposes an explainable machine learning (ML) -based framework for modeling the spatial distribution of earthquake probability across the entire Anatolian Plate. To achieve this, four different tree-based ML model—Random Forest (RF), Extra Trees, Extreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM)—were comparatively evaluated, and a total of 11 spatial variables were incorporated into the classification process. Notably, geodetic strain derived from Global Navigation Satellite Systems (GNSS) was integrated into spatial earthquake probability modeling for the first time, providing a dynamic representation of crustal deformation and demonstrating its significant influence on the model’s decision-making process. Earthquakes with a moment magnitude (Mw) ≥ 4.0 were used for model training, and the dataset was randomly split into 70% training and 30% testing subsets. Model performance was assessed using Accuracy, F1 Score, AUC, and Cohen’s Kappa metrics, and statistical differences between models were tested using the McNemar test. The best-performing model, RF, was interpreted using the SHapley Additive exPlanations (SHAP) method, which clarified the decisive influence of especially proximity to faults, peak ground acceleration (PGA), and geodetic strain, etc., on the model’s decision-making process. The resulting spatial patterns were found to align with major tectonic structures, and the proposed approach presents a interpretable framework that can support seismic hazard assessment in other regions.