Predicting children's energy expenditure during physical activity using deep learning and wearable sensor data.

Area of Science:

• Pediatric exercise science
• Biomedical engineering
• Machine learning applications

Background:

• Accurate assessment of children's energy expenditure is crucial for understanding physical activity (PA) levels and associated health outcomes.
• Triaxial accelerometry offers a promising, non-invasive method for monitoring PA, but its accuracy in children requires further validation.
• Metabolic equivalents (METs) are a standard measure of energy expenditure during physical activities.

Purpose of the Study:

• To evaluate the effectiveness of machine learning models in predicting children's energy expenditure (METs) from accelerometer data.
• To determine the optimal sensor placement (waist, ankle, or wrist) for accurate PA assessment in children.
• To investigate the impact of different physical activities on MET values in children.

Main Methods:

• Twenty-eight healthy children (8-11 years) performed various activities, including sedentary tasks, walking, running, football, and cycling.
• Accelerometers (GENEActiv) were worn on four locations (non-dominant wrist, dominant wrist, dominant waist, dominant ankle) to collect movement data.
• Breath-by-breath calorimetry was used as the criterion measure for energy expenditure, with MET values calculated.

Main Results:

• Machine learning models achieved up to 90% accuracy in predicting METs from accelerometer data.
• Sensor placement on the dominant wrist or ankle demonstrated superior predictive performance compared to other locations.
• Observed MET values ranged from 1.2 (seated play) to 4.1 (running at 6.5 kmph^-1).

Conclusions:

• Novel machine learning models can accurately predict physical activity energy expenditure (METs) in children using accelerometry.
• Dominant wrist and ankle are the most effective sensor placements for reliable PA monitoring in pediatric populations.
• These findings support the use of machine learning with optimized accelerometer placement for objective PA assessment in children.