Analyzing daily behaviours from wearable trackers using linguistic protoforms and fuzzy clustering

The proliferation of low-cost wearable trackers are allowing users to collect daily data from human activity in a non-invasive way and outside of laboratory environments. Exploiting these data properly enable the supervision and counseling from experts remotely; however, extracting key indicators from the long datastreams is hard, often based on statistical metrics or clustering from raw data which lack interpretability. To solve it, we propose an interpretable definition of key indicators by means of linguistic protoforms which include fuzzy temporal processing and fuzzy semantic quantification. Moreover, we use the protoforms defined by experts to evaluate the source datastream in order to provide a straightforward description of the daily activity of users. Finally, the degrees of truth of each protoform are analyzed using a fuzzy clustering method to provide an interpretable description of the longterm user activity. This work includes a case study where data from a user activity (heart beats per minute and sleep stages) have been collected by a Fitbit wearable device and evaluated by the proposed methodology.