Machine learning-driven prediction of axial capacity of concrete-filled steel tubes columns with built-in bamboo or timber cores

Studies on concrete-filled steel tube columns with built-in bamboo or timber cores (CFST-BTC) not only optimize the mechanical properties of conventional concrete-filled steel tube (CFST) columns but also reduce concrete usage and carbon emissions , enhancing structural sustainability . This study establishes a CFST-BTC database of 271 specimens under axial compression and evaluates various machine learning (ML) models for predicting their ultimate bearing capacity ( N cu ). Six mainstream ML algorithms such as ANN , LightGBM, Gradient Boosting Regressor , CatBoost, XGBoost and AdaBoost were assessed for training and predicting N cu of CFST-BTC. Gradient Boosting Regressor and ANN effectively predicted the experimental results, achieving R² values of 0.9988/0.9985 for the training set and 0.9964/0.9959 for the test set, significantly outperforming traditional analytical models. Based on SHAP analysis, key parameters were identified, nonlinear relationships were revealed, and the user-friendly analytical equation was formulated based on ANN . The proposed equation demonstrated strong predictive performance, achieving an R² value of 0.9609 and an RMSE of 156.35. Although these results are slightly inferior to those of the six ML surrogate models , they surpass the explicit equation constructed using multiplication (R² = 0.9297) and other traditional analytical models. The user-friendly equation suggests that the optimal substitution ratio ( w ) of bamboo or timber core replacing the core concrete in CFST-BTC is 36.5 %, providing guidance for the practical application of CFST-BTC columns in engineering. Meanwhile, the suggested explicit equation effectively satisfies the boundary conditions at w = 0 and w = 1, ensuring reliable predictions for ultimate bearing capacity of both CFST columns and bamboo or timber-filled steel tube (BTFST) columns without computational inconsistencies at the boundary conditions. • The concept of concrete-filled steel tubes columns with built-in bamboo or timber cores was established. • A CFST-BTC database of 271 specimens under axial compression were presented. • Boundary conditions were introduced when six machine learning algorithms were predicting axial capacity of CFST-BTC. • A user-friendly equation for a targeted configuration that is superior to the multiplicative configuration was proposed.