Identifying Key Features Associated with Excessive Fructose Intake: A Machine Learning Analysis of a Mexican Cohort

Background: Excessive fructose intake has been linked to adverse metabolic outcomes, yet few studies have comprehensively described the clinical, behavioral, and nutritional patterns associated with different intake levels using machine learning. Methods: In this study, unsupervised and supervised algorithms were applied to a healthy Mexican cohort to examine features related to high fructose consumption, defined as intake above 25 g per day. Results: K-Means clustering identified three distinct profiles, with one subgroup showing less favorable anthropometric, biochemical, and behavioral characteristics. Supervised models, including Extreme Gradient Boosting, Random Forest, and Histogram-based Gradient Boosting, distinguished fructose intake levels with balanced accuracies around 80% and AUC up to 88.1%. Shapley Additive Explanations (SHAPs)-based interpretation highlighted body mass index, triglycerides, sleep duration, alcohol consumption, and anxiety indicators as features most consistently associated with high intake. Conclusions: These findings emphasize the multifactorial nature of fructose consumption and illustrate the utility of machine learning for uncovering dietary and metabolic patterns that warrant further investigation and may guide future nutrition-focused strategies.