Interpretable data‐driven axial bearing capacity analysis of concrete‐filled double‐skin steel tubes

Abstract This study employs machine learning (ML) algorithms to explore the axial bearing capacity of concrete‐filled double‐skin steel tube (CFDST). A dataset containing 210 samples is synthesized, and nine input features are adopted for training and testing six machine learning models. The SHapley Additive exPlanation (SHAP) algorithm is applied to elucidate the prediction processes of ML models and to identify the relationships between the nine input features and the axial bearing capacity of CFDST. Two empirical formulae are employed to compare their performance with ML models. The evaluation metrics comparison reveals that K‐nearest neighbor, artificial neural network, and eXtreme Gradient Boosting models achieve high accuracy, with their R 2 values exceeding 0.98 and MAPE below 7% on both training and testing datasets. Empirical formulae show inferior precision to machine learning models, with their R 2 values below 0.8 and MAPE values above 14%. SHAP analysis indicates that the axial bearing capacity of CFDST is primarily influenced by the diameter of the outer steel tube. In addition, the yield strength of steel in the outer and inner tube, as well as the compressive strength of concrete, are also critical factors. In contrast, the slenderness ratio has a negligible effect when its value is smaller than 3.5. To reduce the impact on axial bearing capacity caused by the absence of core concrete, it is suggested to keep the hollow ratio smaller than 0.5. Besides, the slenderness ratio should be smaller than 8 to alleviate the reduction of axial bearing capacity caused by the bulking of tubes.