Yield strength assessment of partially welded CFST-frame buckling-restrained steel plate shear walls (BRSPSW) through data-driven modeling

The accurate prediction of the yield strength (Py) of L-shaped and square concrete-filled steel tube (CFST) column frame buckling-restrained steel plate shear walls (BRSPSW) walls are important for safeguarding the structures. In this study, six machine learning (ML) models, including Random Forest (RF), Support Vector regression (SVR), XGBoost, Gaussian Process Regression (GPR), AdaBoost, and Artificial Neural Network (ANN), were employed to predict peak strength of the L-shaped- CFST frame-BR composite using 405 dataset from experimental results. The predictive consistency and robustness of each ML model were evaluated using Error distribution analysis. The RF models demonstrated high prediction accuracy with an R² -value of 0.9953, the lowest RMSE (11.81kN) and MAPE (1.347), followed by GPR with R² : 0.9946, RMSE (13.46kN) and MAPE (1.55) at the training phase. However, GPR outperformed others with the highest R² (0.9601) and lowest RMSE (45.16kN) and MAPE (6.081) in the testing phase, demonstrating superior generalization and predictive reliability. SHAP analysis identified yield stiffness (Ky) as the most influential feature, followed by column Area (Ac) and span length (B). This study provides a robust framework for predicting the yield strength of an L/Squre-section- CFST frame BRSPSW using standalone ML models.