Statistical Methods for Wearable Sensor Data
Abstract
Recently, wearable sensors have emerged as promising tools for collecting behavioral data in free-living conditions. For example, physical activity monitors have been used in large population-based studies such as the National Health and Nutrition Examination Survey and the UK Biobank to track participants’ levels of physical activity in their free-living environment. Such large, population-based studies puts forward unprecedented opportunities in exploring demographic, biological, behavioral, and genetic factors associated with physical activity in free living conditions. However, analyzing the large scale behavioral data collected using the wearable devices may warrant special attention for a variety of reasons. First, the device wear times can be highly variable within- and between- individuals over the measurement days, and may be associated with the measurement outcome such as minutes in moderate to vigorous physical activity (informative observation times). Second, study participants may stop wearing the device from a certain measurement day and onwards (censored observations). Third, the early termination of the wearable sensor monitoring may be related to the measured outcome (informative censoring). Fourth, exploration of large number of potential correlates to the wearable sensor measured outcome may necessitate computationally efficient methods. Lastly, rapid developments in the high-resolution sensor technology have demanded more accurate methods for extracting activity intensity features from these data. The overall goal of this study was to develop novel statistical methods to account for the aforementioned challenges in analyzing data from modern wearable devices, scalable for exploring large population level datasets utilizing the state-of-the-art wearable sensor technology.
Subject Area
Biostatistics|Bioinformatics|Epidemiology
Recommended Citation
Song, Jaejoon, "Statistical Methods for Wearable Sensor Data" (2017). Texas Medical Center Dissertations (via ProQuest). AAI10621119.
https://digitalcommons.library.tmc.edu/dissertations/AAI10621119