Dynamic ensemble-learning model for seismic risk assessment of masonry infilled steel structures incorporating soil-foundation-structure interaction

This research is focused on proposing a dynamic ensemble machine-learning (DE-ML) model with the hybrid ability to optimize the hyperparameters by genetic algorithms (GA) and particle swarm optimization (PSO). The purpose of investigation is to estimate seismic probability curves for different kinds of datasets, and decrease the need for modeling and nonlinear analysis of structures. The research investigates a comprehensive dataset of 1536 median of incremental dynamic analysis (M-IDA) curves and 6144 failure probability curves for four demand thresholds, which were determined for the 2-, to 9-story steel moment-resisting frames (MRFs) assuming five soil types with and without soil–foundation–structure interaction (SFSI), four bay lengths, and four implementing conditions of infill masonry walls (IWs). The results show that automated stacked ML (AS-ML) models have superior performance in predicting the measure curves of seismic performance and failure probability (i.e., 97.6 %), while conventional ML models of extreme gradient boosting (XGBoost), gradient boosting machine (GBM), random forest (RF), and LightGBM have a prediction accuracy of 91%–94.2 %. However, proposed DE-ML models have the best-fitted curves and show an accuracy of 99.3 %, while they are compatible with different datasets and case studies. Therefore, they were developed into the GUI to be used for predicting seismic probability curves and risk evaluation of MRFs including IWs and SFSI effects.