Prediction of parameters in the Ibarra–Medina–Krawinkler model for reinforced concrete columns using random forest and active learning

The lumped plasticity model is widely used in the practical modeling of reinforced concrete (RC) columns because of its computational efficiency. However, existing formulations for estimating the backbone curve and cyclic deterioration parameters often fail to accurately predict new data with high variability and necessitate laborious calibrations. To address it, a machine-learning (ML) approach utilizing the random forest (RF) algorithm to predict seven parameters in the Ibarra–Medina–Krawinkler lumped plasticity model for depicting hysteretic response is proposed in this study, where a comprehensive data of 475 RC columns, encompassing diverse material properties, geometric features, and failure modes, is used. In addition, an active-learning framework is integrated to address the limited availability of labeled data in supervised ML tasks, which reduces the exhausted labeling process. The proposed RF surpasses existing research and commonly used ML models regarding coefficient of determination. Additionally, the predicted parameters more accurately simulate the behavior of strength and stiffness degradation in the hysteretic loop of columns than empirical regression formulas. These results will benefit the high-fidelity seismic risk assessments of RC buildings.