Novel ensemble approach to predict the ultimate axial load of CFST columns with different cross-sections

Abstract

Concrete-filled steel tubes (CFST) have been widely used in construction due to their numerous benefits over traditional structural components made of steel or reinforced concrete. The ultimate axial load (UAL) of the CFST column, regarded as the most critical mechanical property, is determined by several constitutive elements, including the material’s mechanical properties and tube cross-sections. Due to the complexity introduced by the interaction of steel tubes and the filling concrete, it is challenging to conduct an analysis of CFST columns using design codes and empirical formulas. As a consequence of this, the principal objective of this study is to use machine learning (ML) methods to predict the UAL of CFST columns. This study mentions six ML models, divided into two groups: the single (or individual) and ensemble ML models. The considered models include Decision Tree, Gradient Boosting, Light Gradient Boosting, Adaptive-Gradient Boosting, Voting Ensemble, and Stacking Ensemble. A database containing 349 test results defining the UAL of rectangular and elliptical CFST columns is compiled to construct the ML models. For model assessment purposes, common quality evaluation criteria, a 10-fold cross-validation approach, and Monte Carlo simulations are utilized. The observed results demonstrate that ensemble ML models outperform single ML models in terms of performance and stability, with the Adaptive-Gradient Boosting model being the best predictive model. In the last step, a partial dependence plot (PDP) analysis is carried out to assess the impact of each input variable and the correlation between variables on the UAL of the CFST columns.