Data-Driven Approach to Predict the Fundamental Period of Steel-Braced RC Frames Using Stacked Generalization Machine Learning Models

Abstract

Abstract The study is directed toward the precise prediction of the fundamental period of steel-braced Reinforced Concrete (RC) Moment-Resisting Frames (MRFs) through the utilization of stacked generalization, an advanced algorithmic ensemble machine learning technique. To facilitate this, a meticulously curated database comprising 17,280 building models has been automated using the ETABS Application Programming Interface (API). The database encompasses both Concentrically Braced Frames (CBFs) and Eccentrically Braced Frames (EBFs) and employs eigenvalue modal analysis to capture the fundamental periods, incorporating diverse bracing configurations and pivotal building parameters. Utilizing SHapley Additive exPlanations (SHAP), the study rigorously scrutinizes influential parameters that affect the fundamental period. The research introduces three stacking ensemble models, with the most effective model employing Random Forest as the meta-model and an ensemble of Extra Trees, Gradient Boosting, XGBoost, LightGBM, CatBoost, and kNN as base models. Hyperparameter tuning was accomplished through Bayesian Optimization, and a thorough sensitivity analysis was conducted. In rigorous evaluations conducted on the test dataset, the proposed model achieved an exceptionally high coefficient of determination (R2) of 0.9889, coupled with an impressively low root mean square error (RMSE) of 0.056. Further validation through multi-dimensional metrics confirmed the model's robust generalization capabilities. Comparative validation against a few popular building code provisions and research models revealed that the proposed model markedly surpasses these benchmarks in predictive accuracy.