Hip and Wrist-Worn Accelerometer Data Analysis for Toddler Activities.

Area of Science:

• Pediatric physical activity research
• Wearable sensor technology
• Behavioral analysis in toddlers

Background:

• Accelerometry is common for older children's activity but less studied in toddlers (1-2 years).
• Toddlers have unique behaviors like stroller rides and being carried, which are not well understood with accelerometry.
• Existing methods may not accurately capture toddler activity levels due to these unique behaviors.

Purpose of the Study:

• To describe accelerometry signal outputs for nine different behaviors in toddlers.
• To investigate the accuracy of accelerometry in differentiating toddler behaviors, including unique ones.
• To inform better methods for assessing physical activity and sedentary behavior in toddlers.

Main Methods:

• Twenty-four toddlers (13-35 months) wore hip and wrist accelerometers.
• Behaviors including running, walking, crawling, sitting, being carried, and stroller rides were video-recorded and annotated.
• Accelerometer data (hip vertical axis, wrist data, vector magnitude) were analyzed for signal outputs during each behavior.

Main Results:

• Hip vertical axis counts for walking were low (median 49 counts/5s) and lower than being carried (median 144 counts/5s).
• Standing, sitting, and stroller rides showed very low hip vertical axis counts (median ≤ 5 counts/5s).
• Machine learning achieved 89% accuracy in differentiating 'carried' from ambulatory movements using various signal features.

Conclusions:

• Hip vertical axis counts alone may not accurately classify walking as activity or being carried as sedentary in toddlers.
• Advanced machine learning techniques using multiple accelerometry features are promising for accurate behavior recognition in toddlers.
• Further research is needed to refine accelerometry interpretation for diverse toddler behaviors.