# Traditional and Non-Traditional Clustering Techniques for Identifying Chrononutrition Patterns in University Students

## TL;DR

This study compares traditional and non-traditional clustering methods to identify meal timing patterns in university students and finds that they reveal similar core structures.

## Contribution

The study is the first to systematically compare clustering techniques for identifying chrononutrition patterns in university students.

## Key findings

- Five distinct meal timing patterns were identified, including Early, Early–Intermediate, Late–Intermediate, Late, and Late with early breakfast.
- Chronotype was aligned with meal timing patterns, with morning types more common in early clusters.
- Food intake quality varied significantly, with healthier diets observed among early eaters compared to late eaters.

## Abstract

Background/Objectives: Chrononutrition—the temporal organization of food intake relative to circadian rhythms—has emerged as an important factor in cardiometabolic health. While meal timing is typically analyzed as an isolated variable, limited research has examined integrated meal timing patterns, and no study has systematically compared clustering approaches for their identification. This cross-sectional study compared four clustering techniques—traditional (K-means, Hierarchical) and non-traditional (Gaussian Mixture Models (GMM), Spectral)—to identify meal timing patterns from habitual breakfast, lunch, and dinner times. Methods: The sample included 388 Mexican university students (72.8% female). Patterns were characterized using sociodemographic, anthropometric, food intake quality, and chronotype data. Clustering method concordance was assessed via Adjusted Rand Index (ARI). Results: We identified five patterns (Early, Early–Intermediate, Late–Intermediate, Late, and Late with early breakfast). No differences were observed in BMI, waist circumference, or age among clusters. Chronotype aligned with patterns (morning types overrepresented in early clusters). Food intake quality differed significantly, with more early eaters showing healthy intake than late eaters. Concordance across clustering methods was moderate (mean ARI = 0.376), with the highest agreement between the traditional and non-traditional techniques (Hierarchical–Spectral = 0.485 and K-means-GMM = 0.408). Conclusions: These findings suggest that, while traditional and non-traditional clustering techniques did not identify identical patterns, they identified similar core structures, supporting complementary pattern detection across algorithmic families. These results highlight the importance of comparing multiple methods and transparently reporting clustering approaches in chrononutrition research. Future studies should generate meal timing patterns in university students from other contexts and investigate whether these patterns are associated with eating patterns and cardiometabolic outcomes.