Advanced machine learning techniques for predicting mechanical properties of eco-friendly self-compacting concrete

This study evaluates the performance of advanced machine learning (ML) models in predicting the mechanical properties of eco-friendly self-compacting concrete (SCC), with a focus on compressive strength, V-funnel time, L-box ratio, and slump flow. The motivation for this study stems from the increasing need to optimize concrete mix designs while minimizing environmental impact and reducing the reliance on costly physical testing. Six ML models-backpropagation neural network (BPNN), random forest regression (RFR), K -nearest neighbors (KNN), stacking, bagging, and eXtreme gradient boosting (XGBoost)-were trained and validated using a comprehensive dataset of 239 mix design parameters. The models' predictive accuracies were assessed using the coefficient of determination, mean squared error, root mean squared error, and mean absolute error. XGBoost consistently outperformed other models, achieving the coefficient of determination values of 0.999, 0.933, and 0.935 for compressive strength in the training, validation, and testing datasets, respectively. Sensitivity analysis revealed that cement, silica fume, coarse aggregate, and superplasticizer positively influenced compressive strength, while water content had a negative impact. These findings highlight the potential of ML models, particularly XGBoost and RFR, in optimizing SCC mix designs, reducing reliance on physical testing, and enhancing sustainability in construction. The application of these models can lead to more efficient and eco-friendly concrete mix designs, benefiting real-world construction projects by improving quality control and reducing costs. • Advanced ML models, including XGBoost, predict SCC properties with high accuracy. • XGBoost outperforms other ML models with R 2 values up to 0.935 for compressive strength. • Sensitivity analysis reveals cement, silica fume, and superplasticizer significantly boost SCC strength.