Related Experiment Video
Updated: Aug 17, 2025
07:24
A Method for Quantifying Upper Limb Performance in Daily Life Using Accelerometers
Published on: April 21, 2017
12.6K
Machine Learning Models for Weight-Bearing Activity Type Recognition Based on Accelerometry in Postmenopausal Women
Cameron J Huggins1, Rebecca Clarke1, Daniel Abasolo1
1Centre for Biomedical Engineering, School of Mechanical Engineering Sciences, Faculty of Engineering and Physical Sciences, University of Surrey, Guildford GU2 7XH, UK.
Sensors (Basel, Switzerland)
|December 11, 2022
Summary
Machine learning models accurately classify physical activity using hip-worn accelerometers in postmenopausal women. This technology can improve assessment of weight-bearing exercises crucial for skeletal health.
Area of Science:
- Biomedical Engineering
- Physical Activity Recognition
- Osteoporosis Research
Background:
- Hip-worn triaxial accelerometers are common for measuring physical activity and energy expenditure.
- Current methods lack classification for specific skeletal-relevant activities, particularly for osteoporosis risk groups.
Purpose of the Study:
- To evaluate the accuracy of four machine learning models for classifying physical activities.
- To assess binary (standing, walking) and tertiary (standing, walking, jogging) classification in postmenopausal women.
Main Methods:
- Eighty postmenopausal women performed shuttle tests; thirty also used a treadmill.
- Accelerometer data underwent pre-processing, feature extraction (18 features, 9 sets), and analysis with four machine learning models.
- Leave-one-out validation was employed to assess model performance.
Main Results:
- A k-NN Manhattan classifier achieved 99.61% accuracy for binary classification (standing/walking) using basic statistical features.
- For tertiary classification (standing/walking/jogging), the highest accuracy was 94.04% with a k-NN Manhattan classifier using all 18 features.
Conclusions:
- The developed machine learning methods accurately classify weight-bearing activities from accelerometer data.
- These techniques can enhance the characterization of physical activity important for skeletal health in at-risk populations.
