Related Experiment Video
Updated: Mar 7, 2026
09:37
Substantiating Appropriate Motion Capture Techniques for the Assessment of Nordic Walking Gait and Posture in Older Adults
Published on: May 12, 2016
9.3K
Classifiers for Accelerometer-Measured Behaviors in Older Women
Dori Rosenberg1, Suneeta Godbole, Katherine Ellis
11Group Health Research Institute, Seattle, WA; 2Department of Family Medicine and Public Health, University of California, San Diego, La Jolla, CA; and 3Fred Hutchinson Cancer Research Center, Seattle, WA.
Medicine and Science in Sports and Exercise
|February 22, 2017
Summary
Machine learning algorithms accurately detect walking and sedentary behaviors in older women using accelerometer data. These validated tools can analyze existing data for improved physical activity monitoring in this population.
Area of Science:
- Gerontology
- Biomedical Engineering
- Physical Activity Epidemiology
Background:
- Accelerometer data analysis is crucial for understanding physical activity and sedentary behavior.
- Limited research exists on machine learning algorithms for activity detection in older adults during free-living conditions.
Purpose of the Study:
- To develop and validate machine learning algorithms for detecting walking and sedentary behaviors in older women using accelerometer data.
- To address the gap in studies focusing on free-living activity detection in older populations.
Main Methods:
- Developed random forest classifiers using accelerometer and SenseCam data in Study 1 (N=39).
- Validated algorithms in Study 2 (N=222) using accelerometer data from observed walk tests and combined accelerometer-GPS data from free-living conditions.
Main Results:
- Study 1 algorithms achieved 82.2% balanced accuracy for classifying behaviors.
- Study 2 demonstrated 87.9% accuracy for walking prediction with machine learning classifiers.
- High agreement (88.6%) was observed between machine learning classifiers and GPS data.
Conclusions:
- Developed free-living algorithms for walking and sedentary time demonstrate high accuracy and concurrent validity.
- These validated algorithms are suitable for application to existing accelerometer datasets from older women.
- The findings support improved physical activity monitoring in aging populations.
