Interpretable prediction of sulfate resistance in high-performance concrete based on ensemble learning and SHAP analysis

Given increasing resource scarcity and environmental pressures, improving high-performance concrete (HPC) durability in marine environments is crucial for sustainable development. Current sulfate resistance assessment suffers from time-consuming and resource-intensive repetitive experimental testing, while existing machine learning approaches focus on isolated durability factors rather than comprehensive multi-variable marine conditions and most studies lack in-depth analysis of complex interactions within available experimental data. To address these limitations, this study develops an ensemble learning framework that leverages existing experimental data to avoid costly repetitive testing, systematically considers multi-variable interactions including SCM types, replacement levels, and employs interpretable analysis to reveal hidden patterns in sulfate resistance mechanisms. A dataset of 241 records from published studies was constructed and evaluated using five ensemble models with the corrosion resistance coefficient (Kf) as target. The Stack model achieved exceptional accuracy (R²=0.961), significantly outperforming individual models. SHAP analysis revealed that GGBS at 15–25 % replacement synergistically enhances durability when combined with fly ash. A cost-durability Pareto optimization framework yielded three mix designs with clear engineering thresholds, providing practical tools for sulfate-rich marine environments.