Related Experiment Video
Updated: Jul 23, 2025
06:51
Physical Activity Measurement in Children Accepting Table Tennis Training
Published on: July 27, 2022
2.0K
Machine Learning-Based Aggression Detection in Children with ADHD Using Sensor-Based Physical Activity Monitoring
Catherine Park1, Mohammad Dehghan Rouzi1, Md Moin Uddin Atique1
1Interdisciplinary Consortium on Advanced Motion Performance (iCAMP), Michael E. DeBakey Department of Surgery, Baylor College of Medicine, Houston, TX 77030, USA.
Sensors (Basel, Switzerland)
|July 11, 2023
Summary
Researchers developed a machine learning model using wearable sensors to objectively detect physical aggression in children. This technology could offer a practical way to monitor and manage aggressive behaviors in daily life.
Area of Science:
- Pediatric behavior analysis
- Wearable sensor technology
- Machine learning in healthcare
Background:
- Childhood aggression is common and impactful, lacking objective measurement tools.
- Current methods for tracking aggression are subjective and lack real-world applicability.
- Objective monitoring is crucial for effective intervention and management of aggressive behaviors.
Purpose of the Study:
- To investigate the use of wearable-sensor-derived physical activity data and machine learning (ML) to objectively identify physical-aggressive incidents in children.
- To develop and validate an ML model for detecting physical aggression with high temporal resolution.
- To explore the potential of sensor data for remote monitoring of aggressive behaviors.
Main Methods:
- Participants (n=39, ages 7-16) wore waist-worn activity monitors for up to one week, repeated over 12 months.
- Demographic, anthropometric, and clinical data were collected.
- A random forest ML model analyzed 1-minute epochs of sensor data to identify physical aggression.
Main Results:
- The ML model achieved high performance metrics: 80.2% precision, 82.0% accuracy, 85.0% recall, 82.4% F1 score, and 89.3% AUC.
- Vector magnitude (triaxial acceleration) was a key feature distinguishing aggression from non-aggression.
- 132 physical aggression epochs were identified within 872 total epochs.
Conclusions:
- Wearable sensor data combined with ML can objectively detect physical aggression in children.
- This approach shows promise for a practical, remote solution for managing childhood aggression.
- Further validation in larger, diverse samples is recommended for clinical implementation.
