This study presents a machine learning (ML) framework to forecast the nonlinear seismic behavior of three-dimensional (3D) RC moment-resisting frames, utilizing OpenSeesPy for realistic 3D modeling and nonlinear time-history analysis (NLTHA) with 110 far-field ground motions from the PEER database. A dataset of 29,700 samples was compiled, spanning 4-, 8-, and 12-story buildings with varied geometries and material properties. Seismic responses, including maximum drift, inter-story drift, and roof drift, were predicted via algorithms such as Extra Trees Regressor (ETR), Random Forest (RF), Gradient Boosting Regression (GBR), and XGBoost. Feature importance identified building height, width, Housner Intensity (HI), and Acceleration Spectrum Intensity (ASI) as key inputs. For inter-story drift, R2 scores varied by height: for 4-story, ETR (0.9388), RF (0.9289), GBR (0.9226); for 8-story, RF (0.9833), GBR (0.9787), ETR (0.9781); for 12-story, GBR (0.9827), ETR (0.9798), RF (0.9622). For maximum drift, XGBoost achieved 0.9684, ETR 0.9678, and RF 0.9612.