One More Bite?: Inferring Food Consumption Level of College Students Using Smartphone Sensing and Self-Reports

While the characterization of food consumption level has been extensively studied in nutrition and psychology research, advancements in passive smartphone sensing have not been fully utilized to complement mobile food diaries in characterizing food consumption levels. In this study, a new dataset regarding the holistic food consumption behavior of 84 college students in Mexico was collected using a mobile application combining passive smartphone sensing and self-reports. We show that factors such as sociability and activity types and levels have an association to food consumption levels. Finally, we define and assess a novel ubicomp task, by using machine learning techniques to infer self-perceived food consumption level (eating as usual, overeating, undereating) with an accuracy of 87.81% in a 3-class classification task by using passive smartphone sensing and self-report data. Furthermore, we show that an accuracy of 83.49% can be achieved for the same classification task by using only smartphone sensing data and time of eating, which is an encouraging step towards building context-aware mobile food diaries and making food diary based apps less tedious for users.