Evaluation of seismic resilience index by using machine learning based on structural performance levels (case study of a hospital in Iran)

Modern society demands higher seismic resilience in structures. This study applies machine learning models to evaluate the Seismic Resilience Index (SRI) of a hospital building using nonlinear analysis-derived structural performance levels. Rather than proposing new algorithms, it integrates structural damage and recovery assumptions into a system-level resilience metric approximated via ensemble learning. Predictive models using Random Forest (RF) and Extreme Gradient Boosting (XGBoost) assess SRI for a seismic hazard level with a 2475-year return period. For this hazard level, RF predicted an SRI of 0.3015 (R² = 0.974), while XGBoost yielded 0.3658 (R² = 0.986). The near-perfect R² scores demonstrate exceptional variance explanation, with XGBoost performing marginally better. The approach effectively links structural performance parameters to resilience metrics, providing a powerful tool for seismic assessment of essential facilities.