Axial strength prediction of seawater sea sand concrete-filled circular FRP tubes under alkaline environment based on ensemble learning algorithms

The rapid development of marine and urban infrastructure led to the extensive studies on seawater sea sand concrete (SWSSC) filled fiber reinforced polymer (FRP) / steel tubes. The material property of FRP is a function of fiber orientations, and the confinement of SWSSC by FRP tubes enhances its axial load carrying capacity. And therefore, its strength prediction is very challenging because of various FRP layouts, surrounding harsh environment and complicated failure modes. The existing empirical models do not consider effects of surrounding seawater environment under elevated temperature. Therefore, this study concentrates on evaluating the axial capacity of two types of physical models: (1) SWSSC filled circular FRP tubes immersed in seawater environment (P_u1), and (2) SWSSC filled FRP-steel-FRP circular tubes (P_u2), with the help of machine learning (ML) algorithms namely extreme gradient boosting (XGBoost), support vector machine (SVM) and artificial neural networks (ANN). The experimental results of 138 and 120 tested specimens were used for the training and testing of ML models. The models were trained with the best hyperparameters based on grid search and cross validation approach. The models were evaluated with number of statistical indices i.e., R², MAE and RMSE accompanied with visual comparison of trend line between experimental and predicted values and predicted to experimental ratios. The ML models were graded on the basis of performance as XGBoost > ANN > SVM for the Model 1, whereas for the Model 2, the order was observed as ANN > XGBoost > SVM. Further, SHAP analysis was conducted based on XGBoost to see the influence of input attributes on ML models.