Detecting daily routines of older adults using sensor time series clustering

The aim of this paper is to develop an algorithm to identify deviations in patterns of day-to-day activities of older adults to generate alerts to the healthcare providers for timely interventions. Daily routines, such as bathroom visits, can be monitored by automated in-home sensor systems. We present a novel approach that finds periodicity in sensor time series data using clustering approach. For this study, we used data set from TigerPlace, a retirement community in Columbia, MO, where apartments are equipped with a network of motion, pressure and depth sensors. A retrospective multiple case study (N=3) design was used to quantify bathroom visits as parts of the older adult's daily routine, over a 10-day period. The distribution of duration, number, and average time between sensor hits was used to define the confidence level for routine visit extraction. Then, a hierarchical clustering was applied to extract periodic patterns. The performance of the proposed method was evaluated through experimental results.