Bites‘n’Bits

Abstract

We collect and analyze mobile data about everyday eating occasions to study eating behavior in relation to its context (time, location, social context, related activities and physical activity). Our contributions are three-fold. First, we deployed a data collection campaign with 122 Swiss university students, resulting in 1208 days of food data, 3414 meal occasions, 1034 snacking occasions, 5097 photos, and 998 days of physical activity. Second, we analyzed the collected data and report findings associated to the compliance, snacks vs. meals patterns, physical activity, and contextual differences between snacks and meals. Third, we addressed a novel ubicomp task, namely the classification of eating occasions (meals vs. snacks) in everyday life. We show that a machine learning method using time of day, time since last intake, and location is able to discriminate eating occasions with 84% accuracy, which significantly outperforms a baseline method based only on time.