Concrete-filled steel tubes (CFST) are recognized for their superior structural performance, offering enhanced strength, ductility, and construction efficiency. This enhanced performance is primarily due to the synergistic relationship between the steel casing and the concrete core; the steel imparts confinement to the concrete, thereby augmenting its compressive strength and ductility, while the concrete infill simultaneously restrains the steel tube against local buckling. This study addresses the critical need for accurate prediction of axial compression capacity in diverse CFST columns, spanning both circular and rectangular cross-sections, and encompassing short to slender configurations. A key focus was accommodating a wide range of material strengths, from conventional normal concrete and steel to contemporary ultra-high-performance concrete and high-strength steel. To achieve this, advanced machine learning (ML) algorithms specifically, ExtraTrees, XGBoost, and GradientBoosting, were employed. The hyperparameters of these models were optimized using Bayesian Optimization to maximize predictive efficacy. The developed models were rigorously validated through 10-fold cross-validation, demonstrating high accuracy, with coefficients of determination (R²) achieving up to 0.99 in predicting CFST column failure loads. Comprehensive error and sensitivity analyses were conducted to thoroughly assess their predictive capabilities and robustness. Furthermore, SHapley Additive exPlanations (SHAP) analysis was utilized to elucidate model decision-making processes and identify the most influential input parameters, enhancing interpretability and confidence. This research significantly advances the efficient, reliable, and data-informed design of CFST structures, expanding their applicability and fostering greater industry acceptance, particularly for innovative designs utilizing high-performance materials.

concrete-filled steel tubes, axial compression capacity, machine learning, bayesian optimization, explainable AI, slenderness