Prediction of load-deformation relations for CFDST columns through machine learning methods

This study explores the load-deformation relations of concrete-filled double-skin steel tubular (CFDST) columns under axial compression, utilizing the machine learning predictive techniques. A comprehensive database of the circular CFDST columns was assembled from previous literature, including their load-deformation relationship data. Criteria were established to classify the types of load-deformation curves. Six machine learning algorithms were employed to predict the yield point, ultimate strength point and failure point, thereby reconstructing the load-strain curve for CFDST columns. Various performance metrics were used to assess the accuracy of these machine learning models. The experimental results were compared and analysed across different models. Furthermore, the predicted ultimate compressive strengths were compared and analysed with the results obtained from the application of current codes of practice. The findings revealed that the machine learning models exhibited superior performance in predicting the load-strain relations of CFDST columns, and the support vector regression (SVR) model exhibited a superior fit to the actual experimental results.