<p>This study tackles the challenge of designing high-performance concrete (HPC) mixtures that support global efforts to achieve net-zero carbon emissions. The study employs a three-loop strategy-refining the machine learning (ML) model, tuning its hyperparameters, and optimizing input variables for multi-objective outcome-to merge ML and multi-objective optimization into a sustainable HPC design framework. Six ML models were benchmarked for predicting HPC compressive strength. Extreme Gradient Boosting (XGB) achieved the highest baseline accuracy; its hyperparameters were tuned via Differential Evolution (DE). Then, the Generalized Differential Evolution 3 (GDE3) algorithm optimized the design input parameters to meet multi-objective design targets, yielding a GDE3-DE-XGB hybrid model. By simultaneously targeting compressive strength, cost, and carbon emissions, the optimization generated Pareto fronts that empower decision-makers to effectively assess and balance these competing design objectives. Beyond its predictive capabilities, the study quantified and visualized how each concrete mixture component influences high-performance concrete behavior using advanced model interpretability techniques. Interpretability analyses translated component-level effects into actionable design rules, and the ML-multi-objective optimization framework systematically refines HPC mixtures—reducing cost and carbon emissions while preserving strength—demonstrating the power of data-driven strategies for sustainable concrete design.</p>