A rhythm analysis-based model to predict sedentary behaviors

Sedentary behaviors such as sitting and watching TV are ubiquitous in modern societies. Increases in sedentary time have been linked with an increased risk of obesity, diabetes, cardiovascular disease, and all-cause mortality. While smartphones and wearables can now detect sedentary user behaviors, few computational models exist for predicting when they will occur in future. In this paper, we propose a lightweight model to predict future sedentary behaviors, facilitating prevention rather than reactive interventions. Our models are based on the concept of rhythm analysis, an idea proposed by Lefebvre, which postulates that many human behaviors, the use of public spaces, and many phenomena all follow natural rhythms. Our work focuses on detecting the prevailing rhythms of sedentary behaviors and modeling the cyclical rhythm and linear rhythm in Lefebvre's philosophy using periodic functions (history-free) and linear functions (history-dependent) respectively. A person who lies on his couch at the same time every day is an example of a cyclical rhythm, while a person who lies down in exhaustion after vigorous exercise is an example of a linear rhythm. Our preliminary results from analyzing an existing dataset clearly show that rhythmical sedentary patterns do exist. Cyclical rhythms are more common than linear rhythms, and half-day rhythms, daily rhythms, weekly rhythms, and biweekly rhythms are clearly observed in a test dataset.