Predicting the Porosity of SCM-Blended Concrete Composites Using Ensemble Machine Learning Models

Cement manufacture is a major source of pollution in the environment as it contributes to 5–7% of total CO₂ emissions globally. Quantity of cement in concrete manufacturing can be reduced by using alternative pozzolanic materials known as supplementary cementitious materials (SCMs). SCMs include a wide range of materials, such as fly ash, slag, metakaolin, silica fume, nanosilica, and other materials that are rich in silica and alumina. These materials are added to concrete mixtures in order to influence various properties of the concrete; one of its important properties is porosity as porosity can have a significant impact on the durability and strength of the concrete. To study the influence of various SCMs on the porosity, either an exhaustive set of experiments or soft computing techniques are needed. This paper presents the use of soft computing techniques as ensemble machine learning (EML) models to predict the values of porosity with differing proportions of SCMs in the concrete mix. Random forest (RF), AdaBoost (AdB), and gradient boosting (GB) were the EML models that were developed in this study. Gradient boosting was shown to be the best predictor of porosity, while the random forest model was found to be subpar after the models were examined under model efficiency parameters. For training, the coefficient of correlation (R²), mean absolute error (MAE), and root-mean-squared error (RMSE) were determined to be 0.995, 0.279, and 0.0.341 for GB, respectively, and for random forest, they were 0.979, 0.383, and 0.677, respectively.