Recognition of activities in children by two uniaxial accelerometers in free-living conditions

The aim of this study was to develop a classification procedure for accelerometer data to recognize the mode of children’s physical activity (PA) in free-living conditions and to compare it with an established cutoff method. Hip and wrist accelerometer data with an epoch interval of 1 s were collected for 7 days from 24 girls (age: 10.7 ± 1.7 years) and 17 boys (age: 10.6 ± 1.6 years). Videos were recorded during the same 7 days at several points of time at school and during leisure time. Each second of video data was labeled as one of nine activity classes. A classification procedure based on pattern recognition algorithms was trained with the accelerometer data relating to respective video labels of half of the children and tested against the data from the other half of the children. The overall recognition rate of the classification procedure was 67%. The procedure was able to classify 90% of stationary activities, 83% of walking, 81% of running and 61% of jumping activities. The remaining activities could not be recognized by the main classifier. This study developed a classification procedure based on well-accepted accelerometers and video recordings to recognize children’s PA in free-living conditions. It has been shown to be valid for the activities of being stationary, walking, running and jumping. In contrast to former measurement and analysis procedures, this method is able to determine the modes of specific activities among children. Consequently, the presented classification procedure provides additional information on the PA behavior in children registered by established accelerometers.