A Preliminary Study for Automatic Activity Labelling on an Elder People ADL Dataset

Abstract

One consequence of the aging population is an increase in life expectancy implying greater healthcare needs as well as a serious healthy aging program. So healthy aging is one of the main challenges in the first world nowadays, and as much as possible devices, software and technological solutions applied to measure and improve the quality of life of the elder people are necessary. Recently, we presented a first prototype of an activity monitoring kit, and this study includes the analysis of the dataset gathered after six months of use. Since the wearable devices employed in this monitoring kit have not the automatic activity recognition service available, current work proposes several techniques to label automatically the Time Series (TS) obtained in the experiment. Thus, a new device with the same sensors as the old one plus the automatic activity recognition service available will be used to obtain a new labelled dataset, that will be used to learn a new model using semi-supervised learning to tag the not-labelled dataset.