RSM, SVM and ANN modeling of the properties of self-compacting concrete with natural mordenite-rich tuff and recycled glass

The present paper is based on the prediction and the modeling of slump flow, l-box ratio and compressive strength of self-compacting concrete, containing natural mordenite-rich tuff as cement substitute and recycled glass as a partial replacement of fine aggregate. The study was carried out on experimental data constructed with a central composite design plan using response surface methodology (RSM), support vector machine (SVM) and artificial neural networks (ANN). Three variable process modelings were used for modeling and optimization: fine aggregate replacement from 0% to 50%, water cement ratio variation from 0.38 to 0.5 and cement substitution with natural mordenite-rich tuff from 0 to 30 %. The RSM, SVM and ANN models were evaluated and compared on the basis of the coefficient of determination (R²), adjusted coefficient of determination (R²_adj), mean square error (MSE) and root mean square error (RMSE). The model’s predictions were accurate with the experimental data with an R² close to 1. The results showed that the slump flow, l-box ratio and compressive strength were strongly influenced (p < 0.01) by the chosen design parameters. The models were found to be robust tools to predict and capture the effects of the design parameters. The ANN outperforms all the regression models. The SVM models for slump flow, l-box ratio were more precise in their estimations in comparison to RSM models. However, in terms of compressive strength the RSM model approach was more accurate. The best optimization setting in terms of concrete properties and environmental consideration corresponds to a high tuff and recycled glass content (30% and 50 % respectively) and low W/C ratio (0.38).