Expert-decision and optimization framework for eco-efficient recycled-powder concrete toward blast-resistant structural applications

• Expert ML–DL framework enables eco-efficient, blast-resistant concretes for Net Zero. • RF–PSO/GA models cut error to 7.67 MPa CS & 12.4 kg CO₂/m³, ensuring accuracy. • PCA & ANOVA validate low-carbon strategies, variance explained >60%. • SHAP shows trade-offs: steel fiber ↑strength, LP & GGBFS ↓emissions. • Optimized mixes cut ∼35% CO₂, aligning with SDG 9, 11 & 13 and Paris targets. The demand for concretes that are both structurally resilient and eco-efficient is increasing; however, achieving high mechanical performance while minimizing carbon intensity remains a major challenge. The framework combines machine learning (ML), hybrid metaheuristic optimization, feature engineering, and explainable artificial intelligence (XAI) to model compressive strength (CS) and CO₂ footprint using a compiled experimental dataset. Multiple predictive strategies were benchmarked, including ensemble learning and hybrid optimization using genetic algorithm and particle swarm optimization (RF–GA and RF–PSO). Statistical reliability was evaluated using Augmented Dickey–Fuller diagnostics and heteroscedasticity tests. Heteroscedasticity diagnostics confirm homoscedastic residual behavior for CS but significant variance heterogeneity for CO₂ footprint, highlighting fundamental differences between mechanical and environmental response mechanisms. Multivariate analyses, including correlation assessment, principal component analysis (PCA), and one-way ANOVA, reveal that reinforcement content and curing conditions primarily govern strength development, while clinker substitution strategies dominate emission reduction. PCA shows that the first three components explain approximately 68% of CS variance and 60% of CO₂ footprint variance, and ANOVA confirms statistically significant heterogeneity among material factors (p < 0.001). In terms of predictive performance, RF–GA and RF–PSO demonstrate the most stable generalization, achieving validation MAE = 7.7 MPa for CS and 12.4 kg CO₂/m³ for CO₂ footprint, outperforming standalone RF. XAI-SHAP analysis identifies steel fiber as the dominant strength-enhancing factor, whereas limestone powder and slag are the most influential contributors to CO₂ footprint reduction. Optimized mixtures achieve up to approximately 35% reduction in embodied carbon while maintaining baseline structural adequacy.