A novel algorithm for activity state recognition using smartwatch data

This work presents a novel algorithm for recognizing activity states which are of interest for assessing the general well-being of cancer, frail and elderly patients. Using the novel idea of two-level classification, misclassification due to unwanted hand motion noise, which is a common source of error in wrist-worn sensing systems, is mitigated. The algorithm is verified using data from 20 subjects performing a sequence of related activities. It is shown that the proposed algorithm improves the accuracy value for the “activity state” which includes “sit”, “stand” and “move” by up to 8%.