Predicting shear strength of beam-column joints based on mechanism-embedded feature optimization method

In order to enhance the generalization and stability of machine learning (ML) models for predicting the shear strength of reinforced concrete (RC) beam-column joints, an efficient prediction model for the shear strength of RC beam-column joints was developed based on a novel mechanism-embedded feature optimization (MEFO) method. The potential features were determined first based on the diagonal strut-truss mechanisms of RC beam-column joints, which constructs a mechanical-embedded initial feature subset. Subsequently, a Fisher score ranking algorithm was further developed to obtain the optimal mechanical-embedded features by using the forward selection elimination strategy, which ensures that the selected features are consistent with both mechanical principles and statistical relevance. Finally, an efficient prediction model for shear strength of RC beam-column joints was developed based on the optimal feature subset and the cross-validation strategy. Analysis results demonstrate that the proposed MEFO method not only enhances the robustness of the feature selection process, but also ensures consistent predictive performance across a variety of algorithms. Compared with traditional methods, the proposed MEFO method has satisfied generalization and stability, which improves the mean absolute error (MAE), the mean squared error (MSE), the root mean squared error (RMSE) and the coefficient of determination (R²) of ML models by 24 %, 33 %, 21 % and 5 %, respectively.