Abstract/Summary

Physical activity recognition plays a vital role in the application of wearable sensors in healthcare. This paper explores the capability of machine learning algorithms to recognise activities of healthy elderly adults and people with Parkinson's (PwP) using wearable sensor data. We examined the potential of triaxial accelerometer alone and with gyroscope for activity recognition. We employed a comprehensive study of several features and classifiers for recognising different activities. The random forest algorithm identified physical activities among elderly people and PwP with an accuracy of 92.29% when both accelerometer and gyroscope sensors used at the same time.