Enhancing axial load prediction of CFDST columns using machine learning models

This study investigated the use of advanced machine learning (ML) techniques to enhance the prediction of axial load capacity in Concrete-Filled Double-Skin Steel Tubular (CFDST) columns.A dataset comprising 255 experimentally tested samples was employed to train and evaluate nine different regression algorithms, including the tree-based ensemble methods.Among these, the XGBoost model, optimized using the Hyperopt-TPE algorithm, demonstrated the highest predictive accuracy, with an R of 0.9856 and RMSE of 195.70 kN.The model's interpretability was further strengthened using SHAP (SHapley Additive exPlanations) and PDP (Partial Dependence Plot) analyses, which identified outer diameter, inner and outer tube thicknesses, and concrete compressive strength as the most influential variables.Additionally, the performance of the XGBoost model was benchmarked against established design codes (ACI, EC4, and AISC), revealing its superior accuracy and robustness in capturing complex nonlinear interactions.The findings suggest that ML models, particularly when enhanced with interpretability frameworks, can complement and surpass traditional code-based approaches in structural capacity estimation, offering reliable, transparent, and efficient tools for engineering design and optimization.These results highlight the strong potential of the model for practical integration into structural engineering workflows, where accurate and explainable predictions are essential for safe and cost-effective design decisions.