Predicting Bioactive Compounds in Arbutus unedo L. Leaves Using Machine Learning: Influence of Extraction Technique, Solvent Type, and Geographical Location

What if a machine could predict the antioxidant treasure hidden in a Mediterranean leaf? A new study uses machine learning to unlock the secrets of Arbutus unedo leaves—no crystal ball needed.

The strawberry tree, Arbutus unedo, is more than a pretty Mediterranean shrub—its leaves are packed with bioactive compounds. But how do extraction methods, solvents, and even the island of origin affect what we get? Researchers collected leaves from two Croatian islands, Vis and Mali Lošinj, and subjected them to three extraction techniques: conventional, Soxhlet, and ultrasound-assisted extraction. They used green solvents—water, 70% ethanol, and ethyl acetate—to keep things eco-friendly.

The results were striking. Solvent type was the star player: 70% ethanol extracted the highest levels of phenols, flavonols, and antioxidant power. Geography mattered too—leaves from Vis boasted more total phenolics and condensed tannins than those from Mali Lošinj. Ultrasound-assisted extraction edged out the others, especially for delicate, heat-sensitive phenolics.

But the real magic happened off the lab bench. Machine learning models, using total phenols as a proxy, predicted other bioactives and antioxidant capacity with remarkable accuracy—Decision Tree and Gradient Boosting achieved R² above 0.91. This data-driven approach could turn leaf extracts into a predictable, scalable resource. Yet, the authors caution: further validation is needed before industrial scale-up. For now, it's a promising blend of green chemistry and smart algorithms.

Key Points

• Solvent type dominates: 70% ethanol maximizes bioactives and antioxidant capacity.
• Geographical origin matters: Vis leaves beat Mali Lošinj in phenolics and tannins.
• Machine learning predicts bioactives with R² > 0.91 using total phenols as proxy.