Unified machine learning approach for predicting CFST column axial load capacity

Concrete-filled steel tube (CFST) columns offer improved strength and ductility, but their complex behavior poses challenges for traditional design equations, which can lead to over- or underestimation of their axial load-carrying capacity (ALCC). This study introduces a unified machine learning (ML) framework that utilizes input space modification and five ML algorithms to accurately predict the ALCC of CFST columns across various cross-sectional shapes. Employing a comprehensive dataset of 3094 CFST column specimens, the proposed framework effectively models ALCC within a single framework. Among the ML models, Extreme Gradient Boosting (XGB) achieved the highest accuracy. Results are compared to predictions from design codes (ACI, EC4, AISC, AS/NZS 2327) and empirical equations, highlighting potential discrepancies attributed to safety factors, column length, and limitations inherent in traditional calculation methods. In addition, SHAP value analysis was conducted to reveal the influence of input features on the ML model’s predictions. This study demonstrates the potential of ML-based approaches for robust ALCC prediction of CFST columns.