Predicting Energy Expenditure in Preschool Children Using Accelerometer and Gyroscope Data

Area of Science:

• Biomedical Engineering
• Pediatric Exercise Science
• Wearable Technology

Background:

• Accurate estimation of energy expenditure (EE) in preschool children is crucial for understanding their physical activity levels and metabolic health.
• Wearable sensors offer a promising, non-invasive method for monitoring EE in free-living conditions.
• Previous research has explored accelerometer data for EE prediction, but the added value of gyroscope data remains less clear.

Purpose of the Study:

• To investigate whether integrating gyroscopic data with accelerometer data enhances the prediction accuracy of energy expenditure in preschool children.
• To compare the performance of different sensor configurations (accelerometer-only, gyroscope-only, and dual-sensor) for EE prediction.
• To evaluate various machine learning models for predicting EE using different sensor inputs and wear locations.

Main Methods:

• Thirty-nine preschool children (aged 3 to <6 years) participated in the study.
• Participants wore multiple devices (OPAL, GT9X, GENEActiv) on different body locations while energy expenditure was measured using a portable metabolic unit.
• Fifty-four machine learning models were developed and evaluated using root mean squared error, comparing accelerometer, gyroscope, and dual-sensor data with different algorithms and wear sites.

Main Results:

• The random forest model incorporating both accelerometer and gyroscope data (dual-sensor) demonstrated marginally lower root mean squared error compared to other configurations in most instances.
• While dual-sensor models showed slight improvements, the enhancement in energy expenditure prediction accuracy was minimal across various settings.
• No significant differences were observed in model performance based on the specific EE measure (metabolic equivalents or kilojoules per minute) or wear location.

Conclusions:

• The addition of gyroscope data to accelerometer data provides only marginal improvements in predicting energy expenditure in preschool children.
• Considering the minimal gains and potential challenges in data acquisition and processing for dual-sensor systems, accelerometer-based models are recommended for future research.
• Future studies should focus on optimizing accelerometer-based algorithms for robust and practical energy expenditure estimation in young children.