Predicting the compressive strength of steel slag concrete: A machine learning-based optimization and cost-benefit analysis

The energy-intensive manufacturing of cement results in a substantial environmental footprint, while the prolonged stockpiling of steel slag (SS) poses serious ecological risks. Increasing the utilization of SS as a cement substitute is an effective strategy for reducing the embodied carbon of concrete and mitigating waste disposal challenges. However, conventional experimental approaches are time-consuming and labor-intensive, and struggle to capture complex interactions governing unconfined compressive strength (UCS). To address this issue, a robust machine learning framework is established for predicting the UCS of steel slag concrete (SSC). The optimized categorical boosting model demonstrates superior performance, yielding an R2 of 0.976 and an uncertainty interval coverage of 92.95% on the test set. Model interpretability reveals that mix and curing factors dominate strength development, contributing a cumulative importance of 67.5%, with curing period and water-to-binder ratio identified as the most influential features. Structural equation modeling further confirms their dominant positive effects through both direct and indirect pathways. Finally, cost-benefit analysis indicates that the optimized cementitious system generates a net benefit of 21.26-61.13 (40.98) CNY/t. These findings provide a data-driven strategy for designing high-performance green concrete with balanced mechanical, economic, and environmental performance.