Data-Driven Machine Learning for Predicting Seismic Response and Fundamental Period of Masonry-Infilled Reinforced Concrete Frames

Abstract

Accurate prediction of seismic response and fundamental vibration period (T) is essential for the robust seismic design of reinforced concrete moment-resisting frames, although masonry infill walls are often neglected. This study developed a hybrid machine learning framework using 600 randomly generated 3D low-rise frames modeled in OpenSees with infill walls. Eigenvalue and response spectrum analyses were performed considering 10 input parameters and four outputs: maximum story displacement, inter-story drift, story shear, and T. Ensemble models were optimized using BOA and WOA with 5-fold cross-validation. XGB-BOA achieved the highest accuracy (99.6% training, 98.7% testing), supported by a web-based GUI.