Machine learning-based prediction of total phenolic and flavonoid in horticultural products

Abstract The purpose of this study was to predict the total phenolic content (TPC) and total flavonoid content (TFC) in several horticultural commodities using near-infrared spectroscopy (NIRS) combined with machine learning. Although models are typically developed for a single product, expanding the coverage of the model can improve efficiency. In this study, 700 samples were used, including varieties of shallot, cayenne pepper, and red chili. The results showed that the TPC model developed yielded R 2cal, root mean squares error in the calibration set, R 2pred, root mean squares error in prediction set, and ratio of performance to deviation values of 0.79, 123.33, 0.78, 124.20, and 2.13, respectively. Meanwhile, the TFC model produced values of 0.71, 44.52, 0.72, 42.10, and 1.87, respectively. The wavelengths 912, 939, and 942 nm are closely related to phenolic compounds and flavonoids. The accuracy of the model in this study produced satisfactory results. Therefore, the application of NIRS and machine learning to horticultural products has a high potential of replacing conventional laboratory analysis TPC and TFC.