Boosting the green microwave-ultrasound-assisted extraction of saffron bioactive compounds: Chemometric and machine learning optimization through OMPD-ANN-SVR comparative study

Optimizing the extraction of key bioactive markers (picrocrocin, safranal, crocin) from saffron (Crocus sativus L.) is crucial. This study optimized Microwave-Ultrasound-Assisted Extraction (MUAE) by comparing Optimal Mixture Process Design (OMPD), Artificial Neural Networks (ANN), and Support Vector Regression (SVR), considering solvent mixture (water/ethanol/glycerol) and process variables: extraction time (10–30 min), temperature (20–100 °C), and microwave power (100–500 W). Multivariate analysis elucidated the complex MUAE dynamics, revealing distinct clusters and highlighting the pivotal influence of temperature. OMPD modeling identified a binary water/ethanol solvent system, maximum power, and maximum extraction time as critical, with optimal temperatures varying per compound. Notably, the ANN model significantly outperformed OMPD and SVR in predictive accuracy, yielding the highest predicted maximum values under optimized conditions: 929.62 for picrocrocin, 703.62 for safranal, and 1058.61 for crocin. Experimental validation using HPLC-DAD confirmed the high efficiency of the ANN-derived optimal conditions, revealing high concentrations of picrocrocin, trans-4-crocine GG, trans-3-crocine GG, trans-2-crocineGg, cis-4-crocine GG, and safranal and demonstrating excellent recovery and precision for the target compounds. This work highlights the superior capability of ANN, integrated with multivariate analysis, for efficiently optimizing complex extraction processes and maximizing yields of valuable saffron bioactives.