Probabilistic multi-objective optimization of sustainable volcanic-ash concrete using explainable machine learning and life-cycle risk analysis across 240 scenarios

Volcanic ash (VA) can reduce cement demand, but the replacement that minimizes environmental burden is not necessarily the mixture that best preserves structural efficiency. This study tested Sustainable Concrete made with freshwater and river sand and a laboratory Marine Concrete system made with physically filtered seawater and sea sand at 0–25% VA and 7–90 days, then integrated experiments, age-constrained explainable machine learning, life-cycle cost analysis (LCCA), and uncertainty analysis. Mechanical performance peaked at 20% VA in both systems and declined at 25%; for Sustainable Concrete, 56-day compressive strength decreased from 41.237 to 35.420 MPa. Separate three-by-three nested grouped validation produced pooled out-of-fold R² values of 0.854, 0.765, and 0.766 for compressive, splitting-tensile, and flexural strength, respectively, with root mean square errors (RMSEs) of 2.954, 0.276, and 0.438 MPa. Y-randomization yielded p = 0.0099 for all three targets, indicating non-random predictive structure without eliminating mixture-group uncertainty. Sustainable Concrete with 20% VA had the lowest life-cycle cost per unit strength, 717.0 PKR·m⁻³·MPa⁻¹, whereas the 25% mixture reached 822.6 PKR·m⁻³·MPa⁻¹ despite a 0.650 probability of being the least-cost alternative. The 20% mixture minimized life-cycle cost per MPa in all 240 structural sensitivity scenarios. Within the tested materials, ages, and economic assumptions, 20% VA therefore provided the most defensible balance; the results do not establish a universal optimum or long-term reinforced-marine durability.