Sensor-based physical activity monitoring: Application of machine learning approaches
Sudhir Chandra Sarangi, Yajnaseni Dash · 2019
Long-lived and healthy lifestyles are the means for sustainable physical and mental wellbeing. Physical activity monitoring is one of the key aspects of maintaining a healthy lifestyle. It can be achieved with the help of widely used smartphones containing accelerometers sensors. The current study proposes the applicability of different machine learning methods for physical activity monitoring based on accelerometer sensor data and their classification. In this paper, Radial Basis Function network (RBF), Multilayer Perceptron (MLP), K-nearest neighbors classifier (KNN) and Random Forest decision tree (RFDT) have been employed for the classification task. Among the classification algorithms, RFDT and KNN recognize all activities very well (97.5% and 96.8%, respectively) with less confusion than other algorithms. Based on accuracy and error scores, it is found that RFDT with the lowest root-mean-square error (0.0758) produce better outcomes than other algorithms. This study observed that RFDT has an important role in classifying the physical activities with better accuracy within a short span of time. Further investigation will be carried out to find out the correlation of physical activity monitoring with different diseases.