Abstract:
This study assesses the predictive performance of three gradient-boosting Machine Learning (ML) models, Gradient Boosting Machine (GBM), eXtreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM), in axial load prediction of circular FRP-concrete-steel double-skin tubular columns (hybrid DSTCs). Data from 275 specimens were compiled from 22 publications in the literature to train and test ML models. Input variables consist of the height of column (), outer diameter of the FRP tube (), outer thickness of the FRP tube (), diameter of the inner steel tube (), thickness of the inner steel tube (), tensile strength of the outer FRP tube (), yield strength of the inner steel tube (), and compressive strength of concrete (), with the ultimate axial load () serving as the output variable. Performance of all three gradient ML models was evaluated using statistical measures including coefficient of determination (R2), root mean square error (RMSE), mean absolute percentage error (MAPE), and mean absolute error (MAE) for training and testing datasets. Results indicate that the XGBoost model performed better than the other two gradient Models (GBM and LightGBM) with R2 values of 0.97 on the training data and 0.95 on the testing data. Further analysis of the XGBoost model assessed the relative importance of input features on the output feature.