Ensemble machine learning with Bayesian optimization predicts bioactive extraction from fermented watermelon rind using natural deep eutectic solvents

What if fermented watermelon rind—often thrown away—could become a treasure trove of health-boosting compounds, predicted by AI?

Watermelon rind is usually compost, but a new study turns it into a source of bioactive compounds using a clever combo: natural deep eutectic solvents (NADES), microwaves, and machine learning. First, the rind undergoes solid-state fermentation, which significantly boosts extraction yields compared to non-fermented controls. Then, microwave-assisted extraction with NADES pulls out the goods, while ensemble machine learning models—fine-tuned with Bayesian optimization—predict the outcomes based on four variables: microwave power, temperature, time, and solid-liquid ratio. The models nailed it, with test R² values up to 0.9252 and tiny overfitting (ΔR² < 0.06). Temperature emerged as the star factor, with importance scores between 0.842 and 0.885, followed by solid-liquid ratio. The optimized conditions (62.5 °C, 27.7 min, 300 W, 30 mg/mL) predicted impressive yields: 1.656 mg CE/g flavonoids, 19.80 mg GAE/g phenolics, and 71.27% DPPH activity. Fermentation alone boosted phenolics from 16.9 to 19.1 mg GAE/g, flavonoids from 0.61 to 0.74 mg CE/g, and DPPH from 66% to 75%. Experimental checks confirmed the models' accuracy, with errors under 3%. This green, AI-powered approach offers a sustainable path to valorize agricultural waste for nutraceuticals.

Key Points

• Fermentation boosts extraction yields significantly.
• Ensemble ML models predict TPC, TFC, DPPH accurately.
• Optimal conditions: 62.5°C, 27.7 min, 300 W, 30 mg/mL.