Related Experiment Video
Updated: Mar 26, 2026
08:45
Assessment of Physical Activity Intensity with Accelerometers and Oxygen Consumption
Published on: June 20, 2025
709
Comparative evaluation of features and techniques for identifying activity type and estimating energy cost from
Rohit J Kate1, Ann M Swartz, Whitney A Welch
1Department of Health Informatics and Administration, University of Wisconsin-Milwaukee, Milwaukee, WI 53211, USA.
Physiological Measurement
|February 11, 2016
Summary
Wearable accelerometers objectively assess physical activity. Using more features improves activity identification, while separate energy cost models per activity are more accurate than a single model.
Area of Science:
- Biomedical Engineering
- Sports Science
- Human Movement Analysis
Background:
- Wearable accelerometers provide objective physical activity data.
- Processing time series data from accelerometers is crucial for accuracy.
- Machine learning methods with specific features are emerging for activity assessment.
Purpose of the Study:
- To experimentally compare machine learning techniques and features for physical activity assessment.
- To evaluate methods for identifying physical activity type and estimating energy cost.
- To determine the influence of features and models on accuracy.
Main Methods:
- Collected data from 146 subjects performing eight different physical activities using hip-worn accelerometers.
- Evaluated statistical, distance-based, and discrete time series features.
- Compared various machine learning techniques for activity type identification and energy cost estimation.
Main Results:
- Increased feature utilization significantly improved physical activity type identification.
- The choice of machine learning technique impacted activity identification accuracy.
- For energy cost estimation, specific models per activity outperformed a general model, with features and techniques being less influential.
Conclusions:
- Feature selection and machine learning model choice are critical for accurate physical activity type identification.
- Activity-specific energy cost models yield superior accuracy compared to a unified model.
- Optimizing data processing methods enhances the utility of wearable accelerometers for health and performance monitoring.
