Multi‐objective optimization of phenolic compound extraction from fresh and dried rosemary ( Salvia rosmarinus ) using evolutionary algorithms

Abstract BACKGROUND A practical challenge in rosemary ( Salvia rosmarinus Spenn.) processing is selecting aqueous extraction conditions that balance total phenolic content (TPC) recovery across fresh and dried matrices. This study combined a Box–Behnken design, Folin–Ciocalteu quantification, complete second‐order response‐surface models, and NSGA‐II to identify operating regions and testable compromise conditions within a shared temperature–time–mass domain. RESULTS Temperature was the principal driver of the modeled response, whereas extraction time showed an early plateau. The dried‐rosemary model showed a tighter fit ( R 2 = 0.905; RMSE = 1.93 g GAE kg −1 sample) and broader modeled high‐response regions than the fresh‐rosemary model ( R 2 = 0.818; RMSE = 3.71 g GAE kg −1 sample). The temperature × mass interaction was supported for fresh rosemary ( P = 5.02 × 10 −7 ) but not for dried rosemary ( P = 0.667). The Pareto front revealed matrix‐dependent trade‐offs. Under equal desirability weights, the selected compromise was 90.0 °C, 4.6 min, and 0.50 g, with predicted TPC values of 30.15 g GAE kg −1 sample for fresh rosemary and 24.31 g GAE kg −1 sample for dried rosemary ( D * = 0.663). CONCLUSION The matched framework provided a common basis for balancing aqueous TPC recovery across raw‐material states. The selected condition and neighboring Pareto solutions are model‐based candidates rather than validated operating recommendations. Independent confirmation runs are required to quantify external prediction error, local robustness, and application potential before bench‐ or pilot‐scale transfer. © 2026 Society of Chemical Industry.