Leveraging Embedded Mobile Device Sensors for Detection of Children's Mobile Screen Use: Examining Feasibility and Performance - PROJECT SUMMARY/ABSRACT Measuring mobile screen use accurately has proven difficult due to limitations in current methods. Direct observation can be costly, invasive, and unfeasible for longitudinal assessment of mobile screen use, while self- and proxy-report struggle to sufficiently capture the intermittent bouts of screen use characteristic of child screen use behavior. Current objective measures of screen time, passive sensing applications, are unable to determine who is using the device, which is of particular concern when children are sharing devices with parents and siblings. A strategy to address this critical shortcoming of objective measures of screen time is using built-in sensors on mobile devices to determine who exactly is using the device. Mobile device sensors (i.e., accelerometer, gyroscope, magnetometer, orientation, touch) have been widely used in behavioral biometric authentication to identify the user while the user is interacting with the device. However, this technology has not yet been applied to mobile screen use measurement to advance our current measures of mobile screen use. To address this, Aim 1 will train machine learning models using mobile device sensor data (accelerometer, gyroscope, magnetometer) to distinguish a target child, through which the most salient features for child biometric authentication will be identified. Aim 2 will then evaluate model performance when machine learning models are tested on a different sample of children during a structured screen time protocol. To accomplish these aims, this study will complement and extend a current R01 (R01DK129215) that has an overarching aim to strengthen the measurement of 24-hour movement behaviors in children using raw sensor data (i.e., heart rate, accelerometry) from consumer wearable devices. This project will leverage sensor data already collected as part of the sedentary block of the R01 protocol to address Aim 1 and will recruit a novel sample of children from the free-living portion of the current R01 to address Aim 2. We will use the sensor- tracking application, SensorLog, on the iPads and data from the built-in accelerometer, gyroscope, and magnetometer sensors will be recorded continuously while the child interacts with the device. This project will use advanced analytic techniques, including machine learning, to address these aims. Findings from the proposed project will open a window of opportunity in child screen time research and offer an innovative approach to objective screen time measurement. Long-term, the goal of this project is to improve the longitudinal monitoring of screen time, which can then inform evidence-based guidelines for screen time in children. As an NIH F31 predoctoral trainee completing this project, the following skills will be accrued: implementing a validation study in children, using advanced analytical techniques, and disseminating research findings in peer-reviewed publications and scientific presentations through a cross-disciplinary lens, all of which will be instrumental in becoming an independent and productive research scientist.
