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

Abstract

The proliferation of low-cost wearable trackers is enabling users to collect daily data on human activity in a non-invasive manner and outside laboratory environments. Properly exploiting these data allows for remote supervision and counseling by experts; however, extracting key indicators from the lengthy data streams is challenging, often relying on statistical metrics or raw data clustering lacking interpretability. To address this issue, we propose an interpretable definition of key indicators using linguistic protoforms, incorporating fuzzy temporal processing and fuzzy semantic quantification. Furthermore, we utilize protoforms defined by experts to evaluate the source data stream, providing a straightforward description of users' daily activity. Subsequently, the degrees of truth of each protoform are analyzed using fuzzy clustering methods to offer an interpretable description of long-term user activity. This work includes a case study wherein data from user activity (heartbeats per minute and sleep stages) were collected using a Fitbit wearable device.