Energy Expenditure Estimation for Schoolchildren Using Accelerometers

As societal emphasis on health intensifies, the accurate estimation of energy expenditure (EE) has become crucial in countering obesity, monitoring activity levels, and fostering improved lifestyle habits, particularly amongst adolescents and the elderly. Extant literature presents several limitations: 1) Predominant focus on adults, leaving a research void concerning middle school students. 2) Concentration on everyday activities, largely overlooking typical school-based physical exercises that students regularly engage in. 3) Limited use of sensors, thereby often failing to accurately capture and delineate energy expenditure and movement across varied body parts. This study seeks to bridge these gaps by focusing on the EE assessment during common physical activities performed by middle school students. We employed seven distinct sensors, strategically positioned on various body parts, to record accelerometer, gyroscope, and magnetometer data from 120 middle school students participating in 10 activities, such as dribbling around poles in soccer, basketball shooting, and badminton. This comprehensive approach ensured a more precise capture and estimation of activity dynamics. Utilizing an Enhanced-GRU network structure, we processed the heterogeneous data from multiple sensor locations, achieving a commendable activity classification accuracy of 98.8% and a mean squared error (MSE) of 0.51 METs in EE prediction. Furthermore, an evaluative comparison of single sensor location revealed that the chest, along with left and right hips, yielded the most optimal results.