Processing Accelerometer Data with an Automated Algorithm – an R Package ‘PhysicalActivity’

• Activity Number: 443 - Making an Impact on Physical Activity and Sleep Research by Developing New Statistical Methods
• Type: Invited
• Date/Time: Wednesday, July 31, 2019 : 8:30 AM to 10:20 AM
• Sponsor: Korean International Statistical Society
• Author(s): Leena Choi* and Cole Beck and Zhouwen Liu and Maciej S Buchowski
• Companies: Vanderbilt University Medical Center and Vanderbilt University Medical Center and Vanderbilt University Medical Center and Vanderbilt University Medical Center
• Keywords: accelerometer; processing data; wear and nonwear classification algorithm; mail delivery classification algorithm; physical activity; R package PhysicalActivity

Abstract

Wearable monitors such as accelerometers have been commonly used to assess physical activity. Data recorded by accelerometer are usually expressed in counts per minute of wearing and could include a substantial number of time intervals during which the monitor was not worn, or the monitor was transported or mailed, which should not be counted as "wear" time. Thus, a crucial step when processing accelerometer data is to remove monitor "nonwear" time intervals including mail delivery days. Since datasets from multiple days of monitoring can be large (>1 GB), using an automated algorithm is an obvious choice. An algorithm for classifying accelerometer "wear" and "nonwear" time intervals, known as "Choi algorithm", was developed and its R implementation "PhysicalActivity" package was published. The package was recently updated to improve speed and robustness of existing functions, and to add several new functions such as a mail delivery day classification algorithm. The accompanying Shiny app was also developed to visualize and summarize data collected by accelerometer without working knowledge of R. We will demonstrate the algorithms with examples and discuss its future development.