Nutritional Characteristics of Foods With Addictive Potential: A Machine-Learning Approach

Objectives. To identify nutritional characteristics associated with the perceived addictive potential of commonly consumed foods in the US food supply, the majority of which are ultraprocessed foods (UPFs). Methods. In a demographically diverse sample of US adults (n = 1664; 55.2% female), participants rated the perceived addictiveness of 297 commonly consumed foods (74.4% UPFs). Data were collected through Prolific in June 2024. Machine-learning models identified nutritional predictors of addictiveness using both the 15 variables required on US Nutrition Facts labels and an expanded set of 166 nutrient characteristics from the Nutrition Data System for Research. Results. Models performed comparably and revealed consistent nonlinear associations between nutrient content and perceived addictiveness. Foods higher in carbohydrates, glycemic load, energy density, and fat were rated as more addictive. These nutrient profiles were rare in minimally processed foods but common in UPFs, which frequently exceeded multiple addictive nutrient thresholds simultaneously. Conclusions. This study identifies a nutritional signature linked to perceived addictive potential. Findings provide a data-driven framework for identifying foods most likely to promote compulsive intake and inform policies aimed at creating a healthier, less addictive food environment. (Am J Public Health. 2026;116(7):950-959. https://doi.org/10.2105/AJPH.2026.308500).