Predicting lying, sitting, walking and running using Apple Watch and Fitbit data.

Area of Science:

• Biomedical Engineering
• Human-Computer Interaction
• Wearable Technology

Background:

• Commercial wearable devices collect vast amounts of physiological and motion data.
• Accurate classification of human movement types is crucial for health monitoring and behavioral analysis.
• Existing methods for movement classification often rely on specialized equipment or less accessible technologies.

Purpose of the Study:

• To evaluate the accuracy of commercial wearable devices in predicting various human movement types.
• To assess the efficacy of machine learning models in classifying activities such as lying, sitting, walking, and running.
• To determine the potential of consumer-grade wearables for large-scale movement pattern analysis.

Main Methods:

• 49 participants wore Apple Watch Series 2, Fitbit Charge HR2, and iPhone 6S.
• A 65-minute protocol included sitting, lying, and treadmill-based activities at varying intensities (3, 5, and 7 METs).
• Machine learning models (Support Vector Machines, Random Forest, Rotation Forest) analyzed heart rate, steps, distance, and calorie data.

Main Results:

• Rotation Forest achieved 82.6% accuracy for Apple Watch data; Random Forest achieved 90.8% for Fitbit data.
• Classification accuracies ranged from 72.6% (sitting) to 89.0% (7 METs) for Apple Watch.
• Accuracies for Fitbit ranged from 86.2% (sitting) to 92.6% (7 METs).

Conclusions:

• Commercial wearable devices can predict movement types with considerable accuracy.
• This study serves as a proof of concept for population-level movement classification using readily available technology.
• Further research is warranted to refine these methods and expand their application in health and wellness.