Fig. 1: Food processing and nutrient changes (FoodProX).
From: Machine learning prediction of the degree of food processing
a, b Ratio of nutrient concentrations for 100 g of Sauteed Onion and Onion Rings compared to Raw Onion, indicating how processing alters the concentration of multiple nutrients. All nutrients in excess of at least two orders of magnitude compared to the concentrations found in the raw ingredient are shown in red. c, d We trained FoodProX, a random forest classifier over the nutrient concentrations within 100 g of each food, tasking it to predict its processing level according to NOVA. FoodProX represents each food by a vector of probabilities {pi}, capturing the likelihood of being classified as unprocessed (NOVA 1), processed culinary ingredients (NOVA 2), processed (NOVA 3), and ultra-processed (NOVA 4). The highest probability determines the final classification label, highlighted in a box on the right. The results shown are for an input list of 99 nutrients. Source data are provided in Source Data Figure 1a–d.xlsx.
