A wearable computing platform for developing cloud-based machine learning models for health monitoring applications.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Wearable Technology

Background:

• Wearable sensors offer potential for clinical-grade ambulatory health monitoring.
• Advances enable precise vital sign measurement, but accuracy degrades in unconstrained environments.
• Translating lab-based measurement accuracy to home settings remains a significant challenge.

Purpose of the Study:

• To present a novel wearable computing platform for unobtrusive data collection.
• To introduce a paradigm for continuous machine learning model development and evaluation.
• To ensure robust model performance during the transition from lab to home settings.

Main Methods:

• Developed a wearable computing platform for labeled dataset collection.
• Implemented a continuous development, deployment, and evaluation paradigm for machine learning models.
• Trained activity classification models across two studies, comparing performance in constrained versus unconstrained settings.

Main Results:

• Demonstrated a system for unobtrusive collection of labeled datasets.
• Tracked changes in machine learning model performance from controlled to unconstrained environments.
• Validated the approach for maintaining robust model performance in real-world settings.

Conclusions:

• The proposed platform and paradigm facilitate the transition of wearable sensor data from lab to home.
• Continuous evaluation ensures reliable machine learning model performance for ambulatory health monitoring.
• This work addresses a key challenge in deploying wearable health technologies in unconstrained environments.