Machine learning assisted multi-objective optimisation and multi-criteria decision analysis ranking for sustainable ready-mixed concrete mix design

To support sustainable ready-mixed concrete (RMC) mix design under the trilemma of performance, cost and carbon dioxide emissions (CDE), a machine learning (ML) assisted workflow was developed that couples multi-objective optimisation with multi-criteria decision analysis (MCDA) ranking. A unified flowability index (FL score) was used to fuse slump and fluidity, and supervised ML surrogates were trained to predict the flowability index and 28-day compressive strength. The best-performing models (extreme gradient boosting for compressive strength (R² = 0.970) and random forest for flowability (R² = 0.902)) were embedded in multi-objective particle swarm optimisation to efficiently explore the design space and generate a Pareto set of 107 candidate mixes that maximised strength and flowability while minimising cost and embodied carbon dioxide. The Pareto candidates were further ranked using a hybrid MCDA scheme that combined analytic hierarchy process preferences with entropy-derived objective weights, yielding stable and interpretable decisions (Kendall concordance coefficient of 0.839). Relative to the engineering C30 benchmark, the preferred mix showed a 15.9% increase in 28-day compressive strength, while reducing CDE and material costs by 21.5% and 2.9%, respectively. The FL score increased from −0.309 to 0.236, corresponding to an absolute increase of 0.545. Shapley additive explanations analysis identified the water-to-binder ratio, curing age and superplasticiser dosage as dominant non-linear drivers. The proposed framework provides a practical, scalable decision-support tool for sustainable RMC production.