Publication Details

Abstract

Accurate prediction of the compressive strength of concrete containing Supplementary Cementitious Materials (SCMs) remains challenging due to the complex and nonlinear interactions among mix constituents. The incorporation of materials such as fly ash and ground granulated blast furnace slag significantly influences the hydration process and microstructural development of concrete, resulting in strength characteristics that are difficult to capture using conventional linear prediction models. These nonlinear interactions limit the effectiveness of a traditional empirical equations for reliable strength prediction. To address this issue, the present study explores the application of machine learning–based predictive techniques capable of modeling complex relationships within concrete mix data. Five predictive models representing different learning strategies were investigated: Random Forest, Gradient Boosting, Extreme Gradient Boosting (XGBoost), AdaBoost, and Multiple Linear Regression (MLR). Random Forest employs a bagging-based ensemble of decision trees to enhance prediction stability, while Gradient Boosting and XGBoost iteratively improve model accuracy by learning from previous errors. AdaBoost focuses on difficult predictions by adjusting model weights, whereas MLR was used as a baseline statistical method. The models were developed using a dataset of 1030 concrete mix designs containing SCMs. The dataset was divided into training and testing subsets in an 80:20 ratio to ensure reliable validation. Model performance was assessed using the coefficient of determination (R²), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and percentage deviation. Results indicate that boosting-based models achieved superior predictive performance, with XGBoost producing the highest accuracy (R² = 0.927). The findings demonstrate the strong capability of machine learning techniques in modeling complex nonlinear relationships in SCM-based concrete systems.