ESTIMATION OF MAXIMUM INTERSTORY DRIFT RATIO OFSTEEL MOMENT-RESISTING FRAMES USING MACHINELEARNING

This study aims to develop machine learning (ML) models that can accurately estimate themaximum interstory drift ratio (MIDR) of steel moment-resisting frames subjected to earthquakeground motions. Four boosting ML methods were applied: gradient boosting, extreme gradientboosting (XGBoost), light gradient boosting machine (LightGBM), and categorical boosting(CatBoost). To train and evaluate the models, a comprehensive dataset was generated through ۹۲,۴۰۰nonlinear dynamic analyses of ۱,۸۴۸ steel moment frames with varying structural properties subjectedto ۵۰ ground motions. The selected ground motions had moment magnitudes of at least ۵.۰ and soilconditions classified as Site Class C. The results suggest that all models could reliably predict theMIDR of steel moment frames when tested on unseen data. The LightGBM model achieved the mostaccurate estimations among those considered, with a coefficient of determination (R۲) of ۰.۹۶۱, meanabsolute percentage error (MAPE) of ۰.۱۹۹%, and root mean square error (RMSE) of ۰.۱۶۵% for thetest dataset.

Keywords:

Maximum Interstory Drift Ratio (MIDR) , Steel Moment-Resisting Frame , MachineLearning , Boosting Methods