• The ML framework predicts the full-range CFST short column’s compressive response. • The predictive model trained on a database of 25,000 experimental data points. • The proposed model achieved R 2 = 0.985 (circular) and R 2 = 0.982 (rectangular). • The proposed model reduces prediction COV to 0.001–0.007 vs. 0.11 in codes. This study marks a significant advancement from conventional single-point ultimate strength predictions. Instead, this research introduces a novel machine learning framework capable of capturing the entire full-range compressive response (force-deformation relationship) of circular and rectangular short concrete-filled steel tube columns, which is critical for performance-based design. The framework is built upon a comprehensive database developed by extracting the full-curve compressive response from approximately 2000 experimental programs, creating a robust dataset of nearly 25,000 data points. The framework incorporates feature engineering to process categorical inputs and utilizes Bayesian hyperparameter optimization to ensure model robustness. Notably, the framework supports vectorized batch processing, enabling the instantaneous generation of full response curves without the computational overhead of iterative modeling. The predicted full-form curves exhibit exceptional alignment with experimental results, and key findings show excellent agreement between ultimate strength values from the model’s predictions and the values from the ANSI/AISC 360-22, AS/NZS 2327, and Eurocode 4 standards. Besides, explainable artificial intelligence analysis confirms that the model’s predictions are consistent with fundamental mechanical principles.
Poorahad et al. (Wed,) studied this question.