Resistance of concrete-filled stainless steel tubular columns with novel hybrid XGB–TO model

The axial load capacity of circular concrete-filled stainless steel tubular columns (CFSSTCs) is vital for the safe and efficient design of high-rise buildings and bridge structures. However, existing design codes often yield inconsistent capacities for CFSSTCs with identical properties, causing uncertainty for engineers. To address this issue, a hybrid machine learning model – extreme gradient boosting with Taguchi optimisation (XGB–TO) – to predict the axial load capacity of CFSSTCs is introduced. An experimental database of 138 samples from prior studies was used to train the model. The XGB–TO model demonstrated high predictive accuracy and reliability (coefficient of determination of 0.99 and a root mean square error of 7.85 kN), outperforming eight existing design code equations. Sensitivity analysis with the Shapley additive explanation (Shap) method and partial dependence plots showed that the specimen diameter was the most important variables affecting axial load capacity of the CFSSTC. The axial load capacity of the CFSSTCs increased with higher tube yield stress and larger specimen diameters, and rose sharply when the tube thickness was >10 mm. Shap analysis confirmed the importance of these features. A visual interface and online tool were also developed to integrate the XGB–TO model, facilitating practical, real-time predictions of axial load capacity for engineers.