Modeling and Optimization of Ultrasound-Assisted Extraction of Antioxidants from Yerba Mate (Ilex paraguariensis St....

The leaves of yerba mate, commonly found in the southern region of Brazil, are consumed as teas and contain a variety of antioxidant compounds with nutritional and pharmacological value. This study aimed to modeling and optimizing the extraction process of these compounds using ultrasound, where operational parameters such as temperature (T), ultrasound power (P), and the solvent-to-solid ratio of yerba mate (R) were evaluated, and four responses measured, including extraction yield, total phenolic content (TPC), and antioxidant capacity determined by the FRAP and DPPH methods. The Box-Behnken design (BBD) was initially used to optimize the variables and generate a response surface model. Subsequently, artificial neural networks (ANN) and adaptive neuro-fuzzy inference system (ANFIS), were applied to enhance prediction and optimization of the results. The findings indicated that ultrasound-assisted extraction (UAE) achieved a TPC of approximately 500 mg GAE g⁻¹ and a DPPH activity of ~ 480 μmol Trolox g⁻¹ under optimal conditions. Statistical modeling using BBD allowed for the identification of optimal operational conditions to maximize the studied responses simultaneously. Additionally, the ANN and ANFIS models demonstrated excellent predictive performance, with high coefficients of determination (R² > 0.99) and low prediction errors (RMSE). The ANFIS method proved particularly effective, accurately predicting yields of antioxidants with R² values of 0.993, 0.998, and 0.994 for Yield, TPC, and DPPH, respectively, along with low RMSE values, corroborating its effectiveness. In conclusion, machine learning tools such as ANN and ANFIS are robust and precise alternatives for modeling UAE as well as other complex processes involving the manipulation and optimization of process variables.