Monte Carlo simulations and advanced modeling of bioactive compound yields in red grapes using green natural deep eutectic solvents

This research introduces a hybrid modeling approach to predict the yields of bioactive compounds from red grapes through microwave-assisted extraction utilizing sugar-based natural deep eutectic solvents (NADES). The experiment consisted of 40 runs with different microwave power levels (700–900 W), temperatures (50–70 °C), times (10–30 min), sample masses (0.5–1.0 g), and ramping rates (0.5–2.5 °C/min). The support vector regression with predicted response surface methodology (SVR-PRSM) and extreme random forest (ERF) were developed and validated using Monte Carlo simulations with synthetic datasets of 500 and 5000 samples. With the SVR-PRSM model, the RMSE for the total anthocyanin content (TAC) was 0.15 mg/g and the RPD was 2.82, while the RMSE for the total phenolic content (TPC) was 0.76 mg/g and the RPD was 1.88. In addition, the kernel density estimation confirmed the model's alignment with the actual distribution patterns, and these results demonstrate the potential of using data-driven optimization of green extraction for applications in the nutraceutical and pharmaceutical industries. • A large-scale dataset was generated by Monte Carlo simulation. • The SVR- PRSM predicts anthocyanin and phenolic yields better than ERF. • High accuracy and reliability of green extraction optimized by machine learning. • Kernel density estimation validated model performance for sustainable processes. • The study developed an effective and eco-friendly extraction process for nutraceuticals.