Unveiling overeating patterns within digital longitudinal data on eating behaviors and contexts

Summary: We analyze long-term digital records of eating behaviors and contexts and identify distinct overeating patterns via a supervised + semi-supervised pipeline.

Abstract

We compile multimodal, longitudinal signals (e.g., self-report, passive sensing, temporal context) to characterize overeating behaviors over months. A pipeline combining supervised classifiers, representation learning, and semi-supervised clustering uncovers five stable overeating patterns with interpretable behavioral and contextual signatures. Cross-validated analyses demonstrate consistent separation and meaningful correlations with temporal routines and reported triggers, suggesting actionable targets for personalized interventions.