Multiple mechanical properties prediction of hydraulic concrete in the form of combined damming by experimental data mining

Abstract The application of combined damming with large volume conventional concrete and roller compacted concrete indicates that the development of hydraulic concrete materials embarking on a new stage. However, the large particle size of coarse aggregate and high fly ash content in hydraulic concrete lead to a highly nonlinear relationship between mixture proportion and mechanical properties. This phenomenon will increase the difficulty, workload, and costs of mechanical experiments. To solve these problems, data mining techniques are mostly applied to the prediction of single concrete performance but not multiple mechanical properties. This research compares four data mining models in predicting the mechanical properties of hydraulic concrete in the form of combined damming. These models involve a linear regression model (Bayesian Ridge), an advanced predictive model (Gaussian Processes), a regression tree model (Decision Trees) and an ensemble learning regression model (Gradient Boosting). The performance measures of these techniques are evaluated on the basis of the concrete data from the combined damming engineering. The results show that ensemble regression model (Gradient Boosting) performs higher accuracy, better measures, and stronger robustness than the other three types of prediction models used to predict the mechanical properties of hydraulic concrete. The feature importance of concrete components derived from Gradient Boosting is analyzed and validated by conclusions of previous studies. Therefore, efficient prediction of mechanical properties and rapid mixture proportion designs of hydraulic concrete can be realized and the construction technology of combined damming will be applied widely.