Affiliation:
1. Guangxi Road and Bridge Engineering Group Co., Ltd., Nanning, China
2. Key Laboratory of Disaster Prevention and Structural Safety of Ministry of Education, College of Civil Engineering and Architecture, Guangxi University, Nanning, China
Abstract
Machine learning (ML) has been applied in civil engineering to predict the compressive strength of concrete with high accuracy. In this paper, five boosting ensemble algorithms, i.e., XGBoost, AdaBoost, GBDT, LightGBM, and CatBoost, were used to predict the compressive strength of high-performance concrete (HPC). The models were evaluated using performance indicators such as R2, root mean square error (RMSE), and mean absolute error (MAE). The results showed that the CatBoost model had the highest accuracy with a R2 (0.970) and a RMSE (2.916). The prediction accuracy of the model was increased through hyperparameter optimization, which got a higher with a R2 (0.975) and a RMSE (2.863). Meanwhile, the SHapley Additive exPlanations (SHAP) method was used to explain the output results of the optimal model (CatBoost), which generated explainable insights that further revealed the complex relationship between the prediction model parameters. The results showed that AGE, W/B, and W/C had the most impact on high-performance concrete compressive strength (HPCCS) prediction, which was similar to the results of sensitivity analysis. This study provided a theoretical basis and technical guidance for developing the mix design of a new high-performance concrete (HPC) system. In the future, the interpretable results of the model output should be iteratively checked and validated in the actual laboratory in order to provide guidance for engineering practice.
Subject
Artificial Intelligence,General Engineering,Statistics and Probability
Reference37 articles.
1. Differences in instantaneous deformability of HS/HPC according to the kind of coarse aggregate;Kliszczewicz;Cement and Concrete Composites,2002
2. Krogh A. and Vedelsby J. , Neural network ensembles, cross validation, and active learning, Advances in Neural Information Processing Systems 7 (1994).
3. Modeling the compressive strength of high-strength concrete: An extreme learning approach;Al-Shamiri;Construction and Building Materials,2019
4. Mechanical properties of concrete containing a high volume of tire–rubber particles;Khaloo;Waste management,2008
5. Analysis and classification of heart rate using CatBoost feature ranking model;Dhananjay;Biomedical Signal Processing and Control,2021