Machine learning approach for predicting Fusarium culmorum and F. proliferatum growth and mycotoxin production in treatments with ethylene-vinyl alcohol copolymer films containing pure components of essential oils

Abstract Fusarium culmorum and F. proliferatum can grow produce, respectively, zearalenone (ZEA) fumonisins (FUM) in different points of the food chain. Application antifungal chemicals to control these fungi mycotoxins increases risk toxic residues foods feeds, induces fungal resistances. In this study, a new multidisciplinary approach based on use bioactive ethylene-vinyl alcohol copolymer (EVOH) films containing pure components essential oils (EOCs) machine learning (ML) methods is evaluated. Bioactive EVOH-EOC were made incorporating cinnamaldehyde (CINHO), citral (CIT), isoeugenol (IEG) or linalool (LIN). Several ML (neural networks, random forests extreme gradient boosted trees) multiple linear regression (MLR) applied compared for modeling growth toxin production under water activity (aw) (0.96 0.99) temperature (20 28 °C) regimes. The effective doses reduce rate (GR) by 50, 90 100% (ED50, ED90, ED100) EOCs EVOH ranges 200 >3330, 450 660 >3330 μg/fungal culture (25 g partly milled maize kernels Petri dish), depending EOC, aw temperature. type film EOC significantly affected GR both species ZEA FUM production. Temperature also only proliferatum. EVOH-CIT was most against Usually, when levels increased, mycotoxin medium decreased although some treatments combination with certain values induced Random forest models predicted better than neural networks trees. MLR mode provided worst performance. This first potential study impact that environmental conditions have results suggest innovative packaging systems be promising tools prediction risks associated toxigenic food.