Wearable Sensor and Algorithm for Automated Measurement of Screen Time

Abstract

The human use of electronic displays (television or computers), also known as "screen time," is currently a topic of great interest within behavior medicine and general clinical research. This behavior has been linked with a wide variety of pathologies, including obesity, circadian disruption, sleeping disorders, cardiometabolic disease, and socio-emotional behavior disorders in children. As an alternative to conventional data collection methods, such as self-reported questionnaires or interviews, we present an automated objective method for estimating screen time that makes use of a wearable wrist band containing an optical color sensor. By applying a machine learning model, and using data collected from a custom designed sensor band, we present results from a small study to demonstrate that it is possible to measure screen time exposure using the color sensor alone without the use of an accelerometer. Using data from two users in two different homes under a variety of activities and lighting conditions, we achieved a classification score of AUC=0.90 for television alone, 0.89 for computer alone, and 0.83 for the combination of both devices. As an additional test, we also present sample results from an experiment in a natural environment. These preliminary results are encouraging and are comparable to the accuracy of conventional self-reported methods. Limitations of this method and potential improvements are also discussed..