Related Experiment Video
Updated: Mar 12, 2026
Physical Activity Measurement in Children Accepting Table Tennis Training
Published on: July 27, 2022
Activity Recognition in Youth Using Single Accelerometer Placed at Wrist or Ankle.
Andrea Mannini1, Mary Rosenberger, William L Haskell
11The BioRobotics Institute, Scuola Superiore Sant'Anna, Pisa, ITALY; 2Stanford Center on Longevity, Stanford University, Stanford, CA; 3Stanford Prevention Research Center, Stanford University, Stanford, CA; and 4College of Computer and Information Science and Bouvé College of Health Sciences, Northeastern University, Boston, MA.
This study adapted an activity recognition algorithm for youth using accelerometer data. The enhanced method improves classification accuracy for both adults and youth, enabling broader application in human activity monitoring.
Area of Science:
- Biomedical Engineering
- Human Activity Recognition
- Wearable Technology
Background:
- Current human activity recognition methods using accelerometers are mainly validated on adult data.
- Body-worn accelerometers offer a promising avenue for objective human activity monitoring.
Purpose of the Study:
- To apply and adapt a previously developed activity classification method using accelerometer data for both adult and youth populations.
- To evaluate the performance of the adapted algorithm on diverse age groups.
Main Methods:
- An existing algorithm for wrist-worn accelerometers, developed with adult data, was tested on youth data.
- New features were incorporated to enhance algorithm performance on youth activity patterns.
- Crossed-validation and leave-one-subject-out cross-validation were employed using combined adult and youth datasets.
Main Results:
- The enhanced feature set improved overall recognition accuracy by 2.3% for adults and 5.1% for youth using wrist data.
- Leave-one-subject-out cross-validation achieved accuracies of 87.0% (wrist) and 94.8% (ankle) for adults, and 91.0% (wrist) and 92.4% (ankle) for youth.
- Combined dataset analysis yielded overall accuracies of 88.5% (wrist) and 91.6% (ankle).
Conclusions:
- Activity classification methods validated in adults can be successfully extended to youth.
- Incorporating youth-specific features and data into the training phase significantly improves algorithm performance for this demographic.
- The refined algorithm effectively distinguishes between ambulation and sedentary, gesturing activities, crucial for large-scale surveillance studies.
