Random forest-based physical activities recognition by using wearable sensors

Abstract

Physical activity recognition (PAR) is a topic worthy of attention. In order to improve the practicality of wearable sensors for recognition, in this study, we propose an approach to create a classifier of PAR based on the collected data. At first, we discuss how features extracted from the accelerometer and gyroscope contribute to distinguish different activities, including walking, walking upstairs, walking downstairs, sitting, standing, laying, and also provide an analytical method employed for this purpose. Then, a supervised machine learning method, random forest algorithm, is adopted to create a classifier to recognize physical activities based on the extracted features. Lastly, the performances of the constructed classifier are evaluated and compared with other methods. The performance evaluation shows the classifier trained by random forest algorithm are better than other algorithms, and its overall recognition rate reaches 93.75%. In addition, our approach also has strong potential for applications in smart textiles.