A0335
Title: A Bayesian motif-based clustering method for discovering activity patterns in free-living physical activity data
Authors: Charlotte Wang - National Taiwan University (Taiwan) [presenting]
Sin-Yu Su - National Taiwan University (Taiwan)
Abstract: Identifying and analyzing activity patterns (motifs) derived from free-living physical activity data collected by wearable devices is a complex and underexplored area. Current literature offers limited methodologies for extracting such patterns through functional cluster analysis. To address this gap, a Bayesian nonparametric motif-based clustering method is proposed to discover activity patterns within daily free-living physical activity data. This algorithm initially segments the 24-hour activity curve into small activity curves and applies elastic functional data analysis to compare them. Subsequently, a Bayesian nonparametric motif-based clustering method is used to explore motifs among these activity curve segments. This approach enables the creation of novel digital biomarkers derived from the clustering of activity patterns to facilitate further biomedical investigations. The proposed method's performance and applicability to biomedical research are evaluated using real-world datasets. It is anticipated that this methodology provides a valuable tool for leveraging wearable device-generated physical activity data in studies exploring the intricate relationships between health behaviors and health events
