Related Experiment Video
Updated: Sep 28, 2025
06:37
Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention
Published on: December 15, 2023
4.2K
Human activity recognition of children with wearable devices using LightGBM machine learning
Gábor Csizmadia1, Krisztina Liszkai-Peres2,3,4, Bence Ferdinandy5
1Department of Ethology, Eötvös Loránd University, Budapest, Hungary. csizmadia.gabor.tamas@ttk.elte.hu.
Scientific Reports
|April 1, 2022
Summary
This study developed a reliable machine learning (ML) method for recognizing children's activities using wearable sensors. The Light Gradient Boosted Machine (LGBM) algorithm successfully identified 17 out of 40 activities, showing promise for human activity recognition (HAR).
Area of Science:
- Computer Science
- Human-Computer Interaction
- Pediatrics
Background:
- Human activity recognition (HAR) using machine learning (ML) is crucial for analyzing behavioral data via wearable sensors.
- Developing automated methods to detect children's playful and daily activities is an ongoing research area.
Purpose of the Study:
- To identify a reliable ML method for automatically detecting various playful and daily routine activities in children.
- To evaluate the effectiveness of the Light Gradient Boosted Machine (LGBM) algorithm for HAR in children.
Main Methods:
- Collected activity motion data from 34 typically developing first graders using wearable smartwatches and SensKid software.
- Defined 40 activities for ML recognition and employed a binary classification task with LGBM and threefold cross-validation.
- Utilized the sliding window technique for signal processing to determine optimal window sizes for behavior analysis.
Main Results:
- The LGBM algorithm successfully recognized 17 out of 40 defined activities with Area Under the Curve (AUC) values exceeding 0.8.
- The window size parameter did not have a significant effect on the recognition accuracy.
- The study demonstrated the potential of LGBM as a robust solution for HAR.
Conclusions:
- The LGBM algorithm shows significant promise as a reliable solution for human activity recognition in children.
- The findings support the development of more precise and effective HAR systems for faster, objective human behavioral analysis.
- This research provides a foundation for advanced HAR systems in pediatric behavioral studies.
