Capturing Occupant Routine Behaviors in Smart Home Environment Using Hierarchical Clustering Models

Abstract

Nearly one in four community-dwelling elders are affected by mild cognitive impairment, comprising a major risk factor for the development of age-related mental disorders, such as dementia. As gradual changes in daily routine is a major symptom of cognitive impairment, the longitudinal and unobtrusive monitoring of occupant's activities of daily living (ADL) in a smart home environment can greatly contribute to the early identification and tracking of progression of cognitive impairment. One important metric in measuring ADL patterns is routine uniformity. However, the high level of complexity in ADL patterns and large amount of noise stemming from real life behaviors pose great challenges in achieving this task. This paper proposes a method to quantify the degree of routineness over multiple time resolutions by identifying clusters of similar activities in a data-driven way. To achieve this, we represent the daily activities over a span of several days as an image, where each pixel represents the activity that occurred at the corresponding time of the day. The segments within the image are then grouped together through bottom-up clustering over various levels of hierarchy based on their relative distance to each other. Our approach was able to capture the routineness of the ADL patterns over a long period of time. Results from this study provide a foundation towards quantifying routine patterns and bouts from the daily routine within an elderly person's life with potential significance to early detection of outcomes of clinical interest.