Machine Learning-Based Comparative Assessment of Nano-SiO2, Nano-TiO2, Nano-Al2O3, Nano-Fe2O3, and Nano-CaCO3 in Concrete

Abstract

The incorporation of nanomaterials into concrete offers a promising approach for modifying its mechanical performance; however, the combined effects of nanomaterial type, dosage, mixture characteristics, and curing age make direct prediction of strength development challenging. This study develops a machine-learning-based framework for predicting the compressive strength (CS), splitting tensile strength (STS), and flexural strength (FS) of nanomaterial-modified concrete. A dataset comprising 360 observations was considered, covering five nanomaterial types—Nano-SiO₂, Nano-TiO₂, Nano-Al₂O₃, Nano-Fe₂O₃, and Nano-CaCO₃—at dosages ranging from 0 to 5%, together with cement content, superplasticizer content, and curing age. Random Forest (RF), Extreme Gradient Boosting (XGBoost), Support Vector Regression (SVR), and Artificial Neural Network (ANN) models were developed using an 80:20 training–testing division, with five-fold cross-validation and hyperparameter optimization applied during model development. Model performance was assessed using correlation coefficient, coefficient of determination, RMSE, MAE, MAPE, NSE, AAD, and SI. Among the investigated approaches, XGBoost provided the best testing performance for all three mechanical properties, achieving R2 values of 0.964839, 0.966156, and 0.964994 for CS, STS, and FS, respectively. The corresponding RMSE values were 0.721804, 0.056642, and 0.080656 MPa. Feature-importance analysis indicated that curing age was the dominant predictor, while nanomaterial type and dosage also contributed substantially to the predictions. Explainable-AI analysis using feature importance, permutation importance, and SHAP was employed to improve interpretation of the developed models. The results demonstrate the potential of machine learning for reliable prediction and interpretation of mechanical properties of nanomaterial-modified concrete.