Publications

A CNN distinguishes pre- and post-mainshock seismic traces, while SHAP identifies physically meaningful 30–40 Hz features linked to foreshock activity.

CNNs classify foreshocks and aftershocks with high accuracy, while SHAP reveals that station-specific effects strongly influence model interpretability.

A CNN classifies foreshocks and aftershocks with 99.66% accuracy, while SHAP links its predictions to a narrow 30 Hz band and reveals possible fault-healing patterns.