The inherently unpredictable nature of earthquakes results in diverse structural responses that can lead to irreversible damage. Steel moment-resisting frames (SMRFs) are widely employed in modern building construction because of their ductility and energy dissipation capacity. Nevertheless, these frames have sustained significant levels of damage in several past seismic events, highlighting the need for improved methods of assessing and predicting their performance. One key factor influencing the seismic behavior of SMRFs is the presence of the surrounding soil domain, which can alter force distribution, stiffness degradation, and overall dynamic response, thereby affecting the interstory drift of such frames. In this study, artificial intelligence techniques, specifically, machine learning algorithms, were adopted to predict the nonlinear interstory drift ratios of SMRFs while accounting for soil–structure interaction effects. Three supervised data-driven learning algorithms were implemented using Python programming. To create training and testing datasets, an SMRF with a two-layer soil profile, consisting of a clay deposit overlying a sand layer, was modeled in the OpenSees platform and analyzed under a suite of ground motions obtained from the Pacific Earthquake Engineering Research (PEER) Center database. The predictive performance of three algorithms, including, linear regression, polynomial regression, and ridge regression, was carefully examined. Among these approaches, polynomial regression provided the highest accuracy, yielding a coefficient of determination (R²) of 0.8525, and was therefore identified as the most effective model for predicting the seismic response of SMRFs.