Machine learning reveals parsimonious differential model for myricetin degradation from scarce data

Abstract Accurately describing the degradation of food antioxidants in oil emulsions, bulk oils, and other foods is essential to understanding and controlling the factors that influence emulsion stability, such as antioxidant concentration. Here, we used a machine learning approach to discover a parsimonious differential equation that describes how myricetin degrades in soybean oil. The differential equation is based on a small experimental dataset, but it describes the degradation of myricetin over a wide range of initial concentrations and extrapolates well beyond the learning data. This information could be used to develop food products with improved shelf life. The machine learning approach we used has the potential to be applied to discover governing equations for other complex food systems, particularly those where the underlying dynamics are difficult to capture from limited experimental data.