Sustainable Extraction of High-Value Phytochemicals from Spontaneous Flora Biomass: Integrating NADES Solvents and Machine Learning Within a Circular Biorefinery Framework

The sustainable valorization of spontaneous flora biomass for the recovery of high value-added phytochemicals represents a key opportunity within the circular bioeconomy, yet it remains constrained by the environmental limitations of conventional extraction solvents and the lack of data-driven optimization frameworks. In this study, Natural Deep Eutectic Solvents (NADES) composed of betaine and 1,3-propanediol were designed and applied as bio-based extraction media for the recovery of bioactive metabolites from Artemisia annua L. spontaneous biomass in the context of green extraction and sustainable resource utilization. Two liquid–solid extraction techniques, namely vortex-assisted extraction and ultrasound-assisted extraction, were evaluated. The influence of key process parameters, including the eutectic component molar ratio, water content, solid-to-liquid (S/L) ratio, extraction temperature, and extraction time, was systematically investigated. Results demonstrated that extraction efficiency was strongly dependent on both solvent composition and process conditions, with distinct optimum parameters for different phytochemical classes. Maximum total polyphenol content (52.08 mg GAE/mL) was achieved via ultrasound-assisted extraction at 20 °C for 15 min, using a 1:3 NADES ratio with 40% water dilution and S/L = 1:5, while the highest flavonoid yield (17.34 mg QE/mL) was obtained by vortex-assisted extraction for 45 min using a 1:6 NADES ratio under the same dilution and S/L conditions. To identify extraction conditions associated with improved process efficiency, a hybrid modeling approach combining deep neural networks with the Success-History-based Adaptive Differential Evolution (SHADE) algorithm was employed, enabling high-accuracy prediction of extraction performance across a broad parameter space. The proposed framework demonstrates the feasibility of integrating green solvent design with machine learning-driven process modeling for the efficient valorization of underutilized plant biomass, contributing to the development of resource-efficient, sustainable extraction protocols, consistent with principles of process intensification and resource-efficient extraction strategies.