Related Experiment Video
Updated: Dec 21, 2025
06:51
Physical Activity Measurement in Children Accepting Table Tennis Training
Published on: July 27, 2022
2.3K
Machine learning to quantify habitual physical activity in children with cerebral palsy
Benjamin I Goodlich1, Ellen L Armstrong1,2, Sean A Horan1
1School of Allied Health Sciences, Griffith University, Gold Coast, Queensland, Australia.
Developmental Medicine and Child Neurology
|May 19, 2020
Summary
Activity monitors and machine learning accurately identify physical activity in children with cerebral palsy (CP) using mobility aids. Combining sensor data significantly improved accuracy, aiding clinical evaluation and research.
Area of Science:
- Biomedical Engineering
- Rehabilitation Science
- Wearable Technology
Background:
- Children and adolescents with cerebral palsy (CP) often have impaired mobility, necessitating assistive devices.
- Accurate assessment of physical activity is crucial for monitoring health and treatment efficacy in this population.
- Current methods for assessing physical activity in children with CP may be limited in accuracy and practicality.
Purpose of the Study:
- To evaluate the accuracy of activity monitors and machine learning models in recognizing physical activity in children and adolescents with CP who use mobility aids.
- To determine the optimal placement and combination of sensors for accurate physical activity classification.
Main Methods:
- Eleven participants with Gross Motor Function Classification System (GMFCS) levels III and IV wore tri-axial accelerometers on the wrist, hip, and thigh.
- Participants completed various physical activity trials, including rest, upper-limb tasks, walking, wheelchair propulsion, and cycling.
- Supervised learning algorithms (decision tree, SVM, random forest) were trained on accelerometer data, with model performance assessed using cross-validation.
Main Results:
- Single accelerometer placements achieved classification accuracies ranging from 59% to 79%.
- The random forest model using wrist data showed the highest single-placement accuracy (79%).
- Combining data from multiple sensors significantly improved accuracy, with a wrist and hip model reaching 92% and a wrist, hip, and thigh model reaching 90%.
Conclusions:
- Machine learning models utilizing raw acceleration signals can accurately recognize physical activity behaviors in children and adolescents with CP using mobility aids within controlled settings.
- These models hold potential for assisting clinicians in evaluating interventions and for researchers studying the benefits of physical activity in children with severe motor impairments.
