Related Experiment Video
Updated: Jun 23, 2025
05:41
A Step-by-Step Implementation of DeepBehavior, Deep Learning Toolbox for Automated Behavior Analysis
Published on: February 6, 2020
9.4K
A Trainable Open-Source Machine Learning Accelerometer Activity Recognition Toolbox: Deep Learning Approach
1Department of Health Science, Institute of Sports Science, University of Bern, Bern, Switzerland.
JMIR AI
|June 14, 2024
Summary
This study introduces HumanActivityRecorder, an open-source smartphone tool for accurate human activity recognition. It uses deep learning to achieve 87% accuracy in classifying behaviors, improving scientific research repeatability.
Area of Science:
- Human-computer interaction
- Biomedical engineering
- Machine learning for healthcare
Background:
- Current activity trackers lack the accuracy and open-source nature required for scientific research.
- Existing movement determination software is insufficient for precise scientific applications.
Purpose of the Study:
- To develop an accurate, trainable, and open-source smartphone-based activity-tracking toolbox.
- To create a system adaptable to new behaviors for research applications.
Main Methods:
- A semisupervised deep learning approach was utilized.
- Activity classification was based on accelerometry and gyroscope data.
- The model was trained and validated using both proprietary and public datasets.
Main Results:
- The HumanActivityRecorder achieved approximately 87% accuracy in classifying 6 distinct behaviors.
- The developed algorithm demonstrated superiority over a dimension-adaptive neural architecture model.
- Robustness against variations in sampling rate and sensor dimensions was confirmed.
Conclusions:
- HumanActivityRecorder offers a versatile, retrainable, and accurate open-source solution for activity tracking.
- The toolbox facilitates adaptation to specific research behaviors and enhances scientific study repeatability.
- Continuous testing on new data ensures the ongoing utility and accuracy of the system.
