Experimental data-driven machine learning approach for predicting workability in sustainable concrete using green material replacements

Concrete workability is a critical factor governing the placement, compaction, and durability of fresh concrete, yet it remains less explored in data-driven studies compared to hardened properties. This study presents an experimentally validated machine learning framework for predicting fresh concrete workability, namely Compaction Factor Equivalent (CFE) and Vee Bee Time Equivalent (VBTE), using a newly generated laboratory dataset comprising 300 concrete mixes. The dataset was developed through controlled experiments by systematically varying the water–binder ratio (W/B), aggregate–binder ratio (A/B), type of green material, and replacement percentage, with fly ash and ground granulated blast-furnace slag (GGBS) used as partial cement replacements to promote sustainability, aligning strategies with SDG 9 (Industry, Innovation, and Infrastructure) and SDG 12 (Responsible Consumption and Production). To capture the nonlinear relationships between mix design parameters and workability indicators, three ensemble learning modelsRandom Forest (RF), Gradient Boosting (GB), and Extreme Gradient Boosting (XGBoost)were developed and evaluated. Model performance was assessed using standard statistical metrics, including R², RMSE, MAE, and MAPE. The results indicate that boosting-based models outperform baseline approaches, with XGBoost achieving the highest prediction accuracy for both CFE and VBTE. By shifting the focus from hardened properties to fresh-state performance, this study addresses a critical research gap and demonstrates that ensemble machine learning models, when combined with experimentally generated datasets, can significantly reduce experimental workload while supporting intelligent and sustainable concrete mix design for practical engineering applications.