Data-Driven Seismic Fragility Assessment of Reinforced Concrete Frames Using Near-Fault Accelerograms

Abstract

This study examines the seismic fragility of 4-, 8-, and 12-story Reinforced Concrete (RC) Moment Resisting Frames (MRFs) under near-fault ground motion using an integrated fragility analysis approach. The research performs an exhaustive seismic fragility assessment of 4-, 8- and 12-story reinforced concrete moment resisting frames (RC MRFs) structures by employing 100 near-fault accelerograms (60 pulse and 40 non-pulse recordings), chosen from PEER NGA-West2 dataset and statistically verified. Ground motion parameters (PGA, PGV, Sa(T1), Sa,avg, Tp) extracted from time histories are used to build dynamic response spectra and to train four ML models based on ensemble methods—Random Forest, Gradient Boosting, XGBoost, and Extra-Trees—with 4,800 Incremental Dynamic Analysis data. An event-based validation approach was employed to avoid any form of data leak from the training set to the test set. XGBoost achieves the best prediction performance (R² = 0.971), generating lognormal fragility curves with an area ratio ≥ 0.96 with nearly 98% reduction in computational cost compared to IDA simulations. SHAP analysis reveals that PGV and Tp are two important features of near-fault ground motion, as the collapse prevention median capacity of structures declines by 33% when shifting from low-rise to tall frame structures.