A Methodology Based on Unsupervised Learning Techniques to Identify the Degree of Food Processing

The level of consumption of processed foods is increasing globally, with a consequent impact on long-term human health due to the increasingly common diseases related to the metabolic syndrome. Therefore, to prevent the continuous spread of these diseases, it is a need to accurately identify the degree of processing of different types of food. There are already several systems that provide rules to classify the level of processing. such as NOVA, but their limitations usually require that nutrition experts follow already established criteria to manually carry out the classification. In this work, we propose a methodology based on unsupervised learning to create food clusters that correspond to their level of processing. In this regard, we apply unsupervised machine learning algorithms to generate food clusters based on their nutritional properties. Training data were retrieved from the Food and Nutrient Database for Dietary Studies (FNDDS). The application of the proposed methodology allowed the differentiation of foods according to their level of processing with an acceptable level of performance. In the end, the proposed methodology is applied to another dataset containing nutrition information from a very large list of real recipes in order to estimate the rate of ultra-processed recipes.