An Open-Source, Open Data Approach to Activity Classification from Triaxial Accelerometry in an Ambulatory Setting.

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Machine Learning in Healthcare

Background:

• Accelerometers are widely used in healthcare for monitoring, offering potential beyond basic activity tracking.
• Existing methods often provide limited insights into nuanced patient movement and activity levels.
• There is a need for open datasets and tools to advance accelerometry applications in health.

Purpose of the Study:

• To create an open dataset and source code for processing tri-axial accelerometry data.
• To develop methods for classifying patient activity levels and types of movement.
• To enable advanced healthcare monitoring and clinical decision-making.

Main Methods:

• Collected 50 Hz tri-axial accelerometry and ECG data from 23 healthy subjects.
• Subjects performed standardized activities: lying, sitting, standing, walking, and jogging.
• Developed two classifiers: signal processing for high/low activity and a CNN for multi-class activity recognition.

Main Results:

• The high/low activity classifier achieved an F1 score of 0.79.
• The multi-class CNN classifier achieved an F1 score of 0.83.
• The dataset and open-source code are publicly available for research.

Conclusions:

• Behavioral activity classification provides valuable context for health metrics.
• This approach can support the development of clinical decision-making tools.
• Findings contribute to patient monitoring, predictive analytics, and personalized health interventions.