Machine learning algorithms to predict physical activities in the GOTOV data
User Modeling and User-Adapted Interaction Journal published the work of Stylianos Paraschiakos et al. presenting how activities of the elderly could be tracked using wearable sensors and Machine Learning with high accuracy (over 90%).
In this study S. Paraschiakos et al built robust and highly accurate activity recognition (AR) model using varying sensor set-ups and levels of activity description granularity using the publicly available GOTOV dataset. In detail, they show that combining accelerometer data from the wrist and ankle with LARA algorithm could generate highly accurate AR models (>93%) for a 12-class classification problem. Furthermore, in this study they demonstrated that their model can be applied accurately in activities performed in free-living conditions and that it can be used to further investigate physical activities in the GOTO intervention study.
As a follow-up to this study S. Paraschiakos et al. used the same GOTOV dataset to estimate Physical Activity Energy Expenditure (PAEE). A pre-print of this work can be found here.