Efficient machine learning models for prediction of concrete strengths

• Published in: Construction & building materials Vol. 266; p. 120950
• Main Authors: Nguyen, Hoang, Vu, Thanh, Vo, Thuc P., Thai, Huu-Tai
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 10.01.2021
• Subjects: Ensemble learning; High performance concrete; Multi-layer Perceptron; Support vector machine; Tree-based algorithms; Tree-based algorithms; High performance concrete; Support vector machine; Ensemble learning; Multi-layer Perceptron
• ISSN: 0950-0618; 1879-0526
• DOI: 10.1016/j.conbuildmat.2020.120950

•Hyperparameters are optimally selected by utilising random search methodology based on highly efficient implementation.•Handling missing data by using the mean of the available data significantly increase predictive performance.•The trained models based on GBR and XGBoost perform better than those of SVR and MLP.•Significant improvement of the current approach in terms of both prediction accuracy and computational effort. In this study, an efficient implementation of machine learning models to predict compressive and tensile strengths of high-performance concrete (HPC) is presented. Four predictive algorithms including support vector regression (SVR), multilayer perceptron (MLP), gradient boosting regressor (GBR), and extreme gradient boosting (XGBoost) are employed. The process of hyperparameter tuning is based on random search that results in trained models with better predictive performances. In addition, the missing data is handled by filling with the mean of the available data which allows more information to be used in the training process. The results on two popular datasets of compressive and tensile strengths of high performance concrete show significant improvement of the current approach in terms of both prediction accuracy and computational effort. The comparative studies reveal that, for this particular prediction problem, the trained models based on GBR and XGBoost perform better than those of SVR and MLP.