Data-driven classification of packaged products commercialized in Uruguay: Insights for the debate on processed food classification systems.

Abstract

The classification of processed foods plays a pivotal role in shaping public health policies and research. While systems such as NOVA categorize foods based on processing level, they have been criticized for definitional ambiguity and for conflating technological, nutritional, and commercial dimensions. This study aimed to contribute to the ongoing debate by applying a data-driven, unsupervised clustering approach to a comprehensive database of 7284 packaged food and beverage products available in the Uruguayan market. Products were grouped based on nutritional composition and food additive disclosure using non-hierarchical clustering based on Gower distances. Nineteen distinct clusters were identified, each characterized by unique nutrient and additive profiles. Principal Component Analysis and heatmap visualizations revealed substantial heterogeneity in formulation both within and across traditional product categories and subcategories. Clusters differed markedly in their use of specific additives, such as acidity regulators, antioxidants, and humectants, demonstrating that additive combinations reflect distinct technological strategies. Notably, additive disclosure did not consistently align with nutrient profiles, reinforcing the multidimensional nature of food formulation. These results call for more granular classification approaches that move beyond prototypical products within categories and explicitly integrate both nutrient profiles and additive disclosure. Developing multidimensional indexes of food formulation could enhance risk assessment and clarify how specific formulation strategies relate to health outcomes, ultimately enabling more precise and effective public health nutrition policies.