Using machine learning models to predict the quality of plant-based foods
Christabel Tachie, Nii Adjetey Tawiah, Alberta N. A. Aryee
Abstract
Plant-based foods (PBFs) are considered healthy, especially minimally processed whole foods, fruits, whole grains, and legumes while highly processed PBFs, maybe less nutritious. Educating consumers on food quality will help to guide their choices. This study aimed at estimating and predicting the nutrient quality of PBFs based on their Nutri-score and micronutrients. The NHANES (2017–2020) data shows the output for foods consumed in the US and their nutrient composition based on a 24-h recall. Though the Nutri-score label has been used to discriminate food quality it still needs to be implemented in most countries and it computes mostly macronutrients with less consideration for micronutrients which contributes to product quality. ML methods used in this study combine the Nutri-score grade and micronutrient in predicting food quality. The FNDDS data of PBFs for 2017–2020 of PBFs were split into training (n = 300) and test (n = 74) datasets. Eight ML models were used to predict the Nutri-score and the Nutri-score grade of PBFs. The random forest (RF) and light gradient boost model (LightGBM) performed best with accuracy and coefficient of determination (R2) scores of 0.88 and 0.96, respectively while DT had the least scores in predicting the Nutri-score grade (0.81) and Nutri-score (0.93). These results suggest that ML methods can effectively predict PBF quality.