Comparative evaluation of hybrid LSTM-Based Models for predicting bioactive compound contents and antioxidant activity in microwave-assisted extraction from carrots using natural deep eutectic solvents

Natural deep eutectic solvents (NADES) are green solvents that, when used in combination with microwave-assisted extraction (MAE), significantly improve the efficiency and sustainability of bioactive compounds extraction. In this study, three hybrid models, Long Short-Term Memory-Long Short-Term Memory (LSTM-RSM), Long Short-Term Memory-Bayesian Response Surface Methodology (LSTM-BRSM), and LSTM-Box Behnken, were used and compared for predicting bioactive compounds extraction from carrots using MAE with NADES. An experiment involving 40 runs examined five critical parameters: microwave power (400–600 W), temperature (50–70 °C), extraction time (10–40 min), sample mass (0.5–1.0 g), and lactic acid concentration (1–2 mol/L). In addition to total phenolic content, total flavonoid content, and 1,1-diphenyl-2-picrylhydrazyl radical (DPPH^●) scavenging activity, the models were tested for their predictive capabilities. According to the sensitivity analysis, microwave power dominates phenolic content, sample mass dominates antioxidant activity, and temperature dominates flavonoid extraction. For the first time, this study presents a novel and sustainable approach for extracting bioactive compounds from carrots by combining NADES with MAE, significantly reducing extraction time, minimizing the use of toxic solvents, and enhancing overall efficiency. In addition to this green extraction technique, three novel hybrid LSTM-based predictive models—LSTM-RSM, LSTM-BRSM, and LSTM-Box Behnken—are introduced for the first time. Among them, the LSTM-Box Behnken model demonstrated excellent predictive ability (R² > 0.99), highlighting the synergy that exists between advanced modeling techniques and environmentally friendly extraction technologies in the process of enhancing the recovery of bioactive compounds.