Explainable machine learning driven axial strength investigation of FRP tube infilled with UHPC

This study employs machine learning (ML) techniques to investigate the compression performance of FRP tube infilled with UHPC. Experimental results of 232 axial compression tests were collected for the training and testing of six ML models, namely Decision Tree (DT), Random Forest (RF), Extremely Random Trees (ERT), Adaptive Boosting (AdaBoost), Gradient Boosting Decision Tree (GBDT), and eXtreme Gradient Boosting Trees (XGBoost). Based on five regression performance metrics, a comparative analysis was carried out on the performance of three empirical formulas and the top three best-performing ML models. The findings showed that both the GBDT and XGBoost models exhibited superior performance, with R² values exceeding 92 % on a benchmark dataset of 214 samples, whereas the empirical formulas performed poorly. Moreover, the Shapley additive explanations (SHAP) method was utilized to interpret the contribution of each input feature to the prediction process of ML models, thus obtaining the correlation ranking between each feature and ultimate strength of UHPC confined with FRP. Notably, the compressive strength of UHPC and the thickness of the FRP tube were identified as the most influential variables, followed by the specimen height. In contrast, the type of FRP and the volume content of steel fiber in UHPC had a minimal impact on the ultimate strength. These insights provide valuable references for optimizing the structural design practice of FRP tube infilled with UHPC. © 2026 Institution of Structural Engineers.