Computational gastronomy: data-driven analysis of food and nutrition

Abstract

The project introduces the development and evaluation of machine learning models incorporating both unsupervised learning clustering and supervised learning classification and regression techniques. These models aim to predict food processing levels utilizing nutrient concentration data, with a specific emphasis on NOVA classification labels. The NOVA system categorizes foods into four groups: unprocessed or minimally processed (Label 1), processed culinary ingredients (Label 2), processed foods (Label 3), and ultra-processed foods (Label 4). Given the escalating concerns about the health impacts of ultra-processed foods, this study seeks to contribute a predictive tool that leverages nutrient analysis for accurate classification. Recent evidence underscores the potential health risks associated with the consumption of ultraprocessed foods. In response, our approach incorporates a combination of unsupervised learning clustering and supervised learning classification and regression algorithms trained on comprehensive nutrient datasets. By utilizing nutrient concentrations as input features, the models strive to precisely classify foods into NOVA labels, offering a nuanced understanding of the processing levels. The methodology involves collecting diverse food samples representing different processing levels. Through the careful adjustment of features and training of models using both unsupervised and supervised learning techniques, robust connections are established between nutrient profiles and NOVA classification labels. Stringent evaluation ensures the reliability and generalizability of the models for predicting processing levels in unseen food items. The results of this study are expected to provide a valuable tool for consumers and health professionals alike to evaluate and track the processing levels in commonly consumed foods. By incorporating both unsupervised learning clustering and supervised learning classification and regression techniques, we enhance our understanding of the complex relationships between nutrient composition and food processing. This, in turn, facilitates informed dietary decisions and contributes to efforts aimed at reducing potential health risks linked to the excessive consumption of highly processed foods.