Cohort Discovery from Bed Sensor Data with Fuzzy Evidence Accumulation Clustering
Trevor M. Bajkowski, Noah Marchal, Jamal Saied-Walker, Pallavi Gupta, James M. Keller, Marjorie Skubic, Grant J. Scott · 2023
In numerous settings, humans are observed in a variety of modalities by connected (e.g., wearables) and unconnected sensors (e.g., stand-off sensors). Smart facilities that leverage unobtrusive sensors for longitudinal health monitoring are able to capture rich data regarding resident health in real-time, a particularly important ability in eldercare. When providing eldercare in technology-enabled living environments, using this wealth of data to detect health changes and identify health conditions is a considerable challenge. Herein, we first examine how data collected from unobtrusive, hydraulic bed sensors can be decomposed into restlessness, respiration, and ballistocardiogram features and subsequently used in clustering algorithms. We apply fuzzy Evidence Accumulation Clustering (Fuzzy-EAC) to the results of clustering these feature sets to discover resident cohorts with similar manifestations of sleep health through unsupervised machine-learning. These cohorts allow clinicians to discover resident links and can provide diagnosis aid based on known medical histories of cohort members. Additionally, the intra-person cluster analysis described here may provide useful insights for enhancing clinical care and improving health outcomes for independent living residents.