Abstract
This study presents the development and evaluation of an Artificial Neural Network (ANN) model for predicting the shear strength of reinforced concrete (RC) beam–column joints subjected to seismic loading. A comprehensive experimental database was compiled from more than 120 RC beam–column joint test specimens reported in the literature and used to train, validate, and test the ANN within MATLAB’s Neural Network Toolbox environment. The model employed the Levenberg–Marquardt backpropagation algorithm, a single hidden layer with an optimized number of neurons, a hyperbolic tangent sigmoid transfer function in the hidden layer, and a linear activation function at the output layer. Input parameters included concrete grade, reinforcement ratio, axial load, and joint geometry, while the output corresponded to joint shear strength. The ANN achieved outstanding predictive performance, with a coefficient of determination (R²) exceeding 0.99 and minimal error metrics (MSE = 0.000105), outperforming multiple regression models and ten widely adopted international design codes. Sensitivity analysis further revealed that reinforcement ratio and axial load were the most influential predictors of joint shear capacity. In addition to numerical prediction, the ANN demonstrated strong generalization capability and robustness across different concrete grades (M25–M40) and design standards. The results highlight the superior adaptability of machine learning compared to conventional design approaches, offering an innovative, data-driven framework for seismic performance assessment. This research contributes to the advancement of performance-based design methodologies by integrating artificial intelligence into structural engineering, paving the way for more accurate, efficient, and reliable seismic safety evaluations of RC joints.