Classifying Physical Activity Levels in Early Childhood Using Actigraph and Machine Learning Method

Abstract

Actigraph is a widely used accelerometer for classifying physical activity levels in children, adolescents, adults, and older people. The classification of physical activity levels on Actigraph is determined through time calculations using cut-point formulas. The study aims to classify physical activity in young children according to the guidelines of the World Health Organization (WHO) using accelerometer data and machine learning methods. The study involved 52 young children (26 girls and 26 boys) aged 4 to 5 years in West Java, with an average age of 4.58 years. Physical activity and sedentary behavior of these early childhood were simultaneously recorded using the Actigraph GT3X accelerometer for seven days. The data from the Actigraph were analyzed using two algorithm models: the decision tree and support vector machine, with the Rapidminer application. The results from the decision tree model show a classification accuracy of 96.00% in categorizing physical activities in young children. On the other hand, the support vector machine model achieved an accuracy of 84.67% in classifying physical activities in young children. The decision tree outperforms the support vector machine in accurately classifying physical activities in early childhood. This research highlights the potential benefits of machine learning in sports and physical activity sciences, indicating the need for further development.