High-performance concrete strength prediction based on ensemble learning

• Four ensemble learning models, AdaBoost, GBDT, XGBoost, and random forest, were used to study. • The effects of the dataset division ratio on model performance were explored through tests. • The model shows superiority in comparison with traditional machine learning models. • The model with the best prediction performance is GBDT. The compressive strength and tensile strength of high-performance concrete (HPC) are important mechanical property indexes. However, the related mechanical tests are time-consuming; therefore, predicting the strength of HPC using available test data is important. In this study, compressive strength and tensile strength tests were conducted on HPC with fly ash and silica fume separately, with fly ash and silica fume together, and with fly ash, silica fume, and polypropylene fiber in triple-blending. Based on the analysis of the test data, the contribution of silica fume to the increase in compressive strength and tensile strength occurred in the early stage of maintenance, whereas the contribution of fly ash to the increase in compressive strength and tensile strength occurred in the late stage of maintenance. Four ensemble learning models, AdaBoost, GBDT, XGBoost and random forest, were used in this study. The optimal data set division ratio was tested to be 8:2. The sensitivity of the input variables was obtained through the model. The best prediction model among the four ensemble learning models established was GBDT, and the GBDT model showed a good performance with other machine learning models.