Publication: Human activity recognition using triaxial acceleration data from smartphone and ensemble learning
3
Issued Date
2018-04-09
Resource Type
Other identifier(s)
2-s2.0-85048863127
Rights
Mahidol University
Rights Holder(s)
SCOPUS
Bibliographic Citation
Proceedings - 13th International Conference on Signal-Image Technology and Internet-Based Systems, SITIS 2017. Vol.2018-January, (2018), 408-412
Suggested Citation
Narit Hnoohom, Sakorn Mekruksavanich, Anuchit Jitpattanakul Human activity recognition using triaxial acceleration data from smartphone and ensemble learning. Proceedings - 13th International Conference on Signal-Image Technology and Internet-Based Systems, SITIS 2017. Vol.2018-January, (2018), 408-412. doi:10.1109/SITIS.2017.73 Retrieved from: https://repository.li.mahidol.ac.th/handle/123456789/45641
Research Projects
Organizational Units
Authors
Journal Issue
Thesis
Title
Human activity recognition using triaxial acceleration data from smartphone and ensemble learning
Other Contributor(s)
Abstract
© 2017 IEEE. In recent years, the use of smartphone sensors in human activity recognition (HAR) has been well studied. Mostly, a smartphone accelerometer has played the main role to solve the problem of HAR. However, the role of a gyroscope is to be explored, both when used alone as well as in combination with an accelerometer. For this purpose, the researchers investigated the role of these two smartphone sensors in human activity recognition. Two ensemble learning based approaches, i.e., majority voting and stacking, to improve recognition performance were presented. Also, the researchers evaluated the roles of the approaches on two body positions using the two ensemble classifiers while recognizing six physical activities, i.e., standing, sitting, laying, walking, walking upstairs, and walking downstairs. From the experimental results, it was shown that in general an accelerometer and a gyroscope complement each other, thereby making the recognition performance higher. Moreover, the ensemble learning based approaches could improve the recognition performance in terms of accuracy to 91.1667 percent.