Identifying Free-Living Physical Activities Using Lab-Based Models with Wearable Accelerometers.

Area of Science:

• Wearable sensor technology
• Human activity recognition
• Biomedical engineering

Background:

• Accurate physical activity monitoring is crucial for health research and interventions.
• Wearable sensors offer a promising avenue for objective physical activity assessment.
• Distinguishing between various physical activities in free-living settings remains a challenge.

Purpose of the Study:

• To classify and model diverse physical activities using supervised lab-based data.
• To develop a model for identifying physical activity in free-living environments.
• To assess the accuracy and characteristics of free-living physical activities.

Main Methods:

• Collected wrist-worn accelerometer data from 152 adult participants in a controlled lab setting.
• Employed Gaussian Mixture Models (GMM) and Hidden Markov Models (HMM) to classify 24 physical activities.
• Validated the best performing model in free-living conditions with 20 participants over 40 sessions.

Main Results:

• Achieved high classification accuracy for physical activities in controlled settings (92.7% GMM, 94.7% HMM).
• Successfully applied the model to identify free-living activities with 80% accuracy.
• Identified a prevalence of stationary and light-intensity activities in 36 out of 40 free-living sessions.

Conclusions:

• Proposes a novel approach for unsupervised free-living activity recognition using lab-based models.
• Demonstrates the utility of wearable sensors for identifying physical activities and estimating energy expenditure.
• Highlights the dominance of lower-intensity activities in typical daily routines.