Fragility Curves and Machine Learning-Based Damage Classification Models from Strong Motion Data Measured During the 2023 Turkey-Syria Earthquake

This paper presents a set of fragility curves and machine learning-based damage classification models developed using strong motion data from the 2023 Turkey-Syria earthquakes. The models rely on post-event performance evaluations of 322 buildings and strong ground motion records measured during the events and use as input ground motion intensity measures/parameters computed directly from the ground motion records, such as Peak Ground Acceleration, Peak Ground Velocity, Cumulative Absolute Velocity, Root-Mean-Square Acceleration, Arias Intensity, Araya-Saragoni Intensity, Housner Spectral Intensity, and Pseudo-Spectral Acceleration. The study explores the damage-predicting capability and the correlations between the aforementioned intensity measures and the observed structural damage. The damage predicting models consist of a set of fragility curves generated using the maximum likelihood method, and a random forest classifier with a feature importance analysis that includes a SHapley Additive exPlanations (SHAP) model. The results show that, for the dataset studied, the Peak Ground Acceleration and Araya-Saragoni Intensity are the most effective damage predictors. The models demonstrate how post-earthquake data can inform seismic performance assessment and enhance risk-informed decision-making.