Estimating Seismic Behavior of Buckling-Restrained Braced Frames Using Machine Learning Algorithms

Over the last few decades, there has been a growing interest in exploring the seismic behavior of buckling-restrained braced frames (BRBFs) as a passive device for dissipating seismic energy. Machine learning (ML) methods of decision forest (DF), artificial neural networks (ANNs), gradient boosting machines (GBM), and LightGBM were used to predict the seismic response of two- to ten-story BRBFs located in soil D. The partial dependence-based features selection method is proposed to increase the capability of methods for the estimation of seismic responses of BRBFs subjected to far-fault ground motions. The results showed that the GBM and DF methods with accuracy of 97.2% and 95.6%, respectively, can be used to predict the seismic response of BRBFs. Therefore, applying the proposed methods can facilitate the response prediction procedures and help designers, while decreasing the total computational efforts.