Classification of Physical Activities Using Wearable Sensors
Taylor & Francis
Intelligent Automation and Soft Computing
An effective measurement of physical activity gives an accurate indication of physical health. This information can be highly useful, particularly in rehabilitation development and personal weight management. In this paper, time domain features are selected for different types of physical activities to produce the best classification in the feature space. We have used wavelet transform based rotation forest classifier to recognize seventeen different types of physical activities. Furthermore, to improve the time and space complexity, we have compared three types of attribute or feature selection methods. Our proposed framework has produced higher classification accuracy of 98% with only 58 features selected by correlation based feature selection (CFS) using tabu search for seventeen different physical activities.