Effects of variability in experimental database on machine-learning-based prediction of ultimate load of circular concrete-filled steel tubes

Abstract This study investigates the performance and robustness of regression machine-learning models in the presence of variability in the experimental database. The main objective of this work is to predict the ultimate load of circular concrete-filled steel tubes. The simulations were designed by combining size of the learning dataset, random realizations and prediction models. The variability (i.e. probability density function of each variable) is propagated to the output response through the regression machine-learning models. Results show that such variability must be considered when training and testing regression machine-learning models. The performance and robustness of the prediction models are presented and discussed. Based on the most robust and efficient model, a prediction equation is proposed for practical use. After conducting a comparison investigation, the performance of the proposed equation is found superior to one of current models. Finally, the proposed equation is implemented in Excel and appended to this paper.