This study presents Eco-StrengthNet, a novel framework that unifies concrete mix–design and machine-learning ensemble tuning into a single triobjective optimisation. Leveraging a stacking ensemble of LightGBM, CatBoost, HistGradientBoosting, MLP, and SVR–with an ElasticNet meta-learner–Eco-StrengthNet integrates with NSGA-II to simultaneously maximize 28-day compressive strength and minimize embodied CO \(_2\) and energy consumption. On a 1 000-sample Sustainable Concrete Mixture dataset, the method achieved RMSE = 10.42 MPa, MAE = 7.36 MPa, and \(R^2 = 0.9905\) on hold-out test data, outperforming six classical and boosting baselines. The knee solution on the Pareto front yielded RMSE = 5.21 MPa with a 5 % improvement in normalized eco-penalty. A comprehensive explainability toolkit–including permutation importance, SHAP analysis, Sobol sensitivity indices, and 2D PDP surfaces–confirmed that curing age dominates strength prediction, while sustainability metrics exert secondary influence. Finally, a Streamlit GUI enables practitioners to explore mix–performance trade-offs in real time. This study demonstrates that integrated model–mix optimisation advances both performance and environmental sustainability in concrete design.
