A comparison of different machine learning algorithms, types and placements of activity monitors for physical activity classification

This study classified physical activities using supervised machine learning (SML) algorithms based on accelerometer measures. The influences of different types, placements, and monitor modalities of the GT3X+ and GT9X have been further analysed. Specifically, 9 healthy participants were recruited to perform 14 activities by wearing GT3X+ and GT9X together at the hip and the thigh, respectively. Four different SML algorithms were utilized and evaluated in the classification of physical activities. The experimental results showed that the performance of the SML algorithms would not be affected by different placements and monitor modalities. Support vector machine performed satisfactorily across all monitor modalities (around 89% accuracy rate). Meanwhile, in both placements of the hip and the thigh, the overall accuracy of the GT9X was not better than that of the GT3X+, and the overall accuracy of the combined mode (two monitors together) was not better than that of the single mode (one monitor). (C) 2020 Elsevier Ltd. All rights reserved.