Predicting Shear Strength of Exterior RC Beam–Column Joints Using Regression and Machine Learning Models

Abstract

Accurately predicting the shear strength of exterior reinforced concrete (RC) beam–column joints remains challenging due to the complex interaction of geometric and material parameters and inconsistencies among existing analytical and code‐based models. In this study, a database of 273 experimentally tested specimens was compiled, incorporating 15 structural and material input parameters. Using the full set of 15 inputs, the interaction regression formulation demonstrated the highest prediction accuracy, representing the optimal model under full‐information conditions. To complement this evaluation, eight analytical and code‐based shear strength models were also examined using their respective input variables, resulting in model‐specific reduced datasets. Among these formulations, the Kassem shear strength model, which includes 13 input parameters, achieved the best agreement with experimental results when expressed through a multiplicative regression form, outperforming the remaining seven models. These findings indicate that the interaction formulation is more effective when comprehensive input data are available, whereas the multiplicative Kassem‐based model provides the most reliable predictions among existing analytical and code‐based approaches. Overall, this dual perspective enables more accurate and context‐dependent estimation of shear strength in RC beam–column joints.