Comparing activity trackers to investigate physical activity
Physical activity is typically measured with questionnaires that are known to have measurement issues, and specifically, overestimate the amount of physical activity of the population. This paper describes the search for a sensor system that is accurately and validly capable to recognise physical activity and its intensity.
Different methods to estimate the intensity of physical activity are compared, and machine learning models are trained to classify the activities. The models are applied to laboratory data of 40 participants who wore four different sensors on five different body parts, while performing various activities (sitting, standing, stepping with two intensities, bicycling with two intensities, walking stairs and jumping). They also kept a diary keeping track of their physical activities, work and travel hours.