Online Detection of Behavioral Change Using Unobtrusive Eldercare Monitoring System

The rapid ageing population is posing challenges to many countries all over the world, particularly in the provision of care to the growing number of elderly who are living alone. Allowing the elderly to age-in-place, i.e., live safely and independently in the comfort of their own homes is a model that can potentially address the resource constraint in health and community care faced by many nations. To make this model a reality and provide appropriate and timely care to the elderly, unobtrusive eldercare monitoring systems (EMS) are being deployed in real homes to continuously monitor the activity of the elderly. In this paper, we study the feasibility of detecting behavioral changes using rudimentary binary sensors similar to the ones used by many commercial EMS, as a trigger for early intervention by caregivers. We propose Online Behavioral Change Detection (OBCD), a scheme to automatically detect behavioral changes using online streaming data from binary sensors. OBCD extends existing changepoint detection methods to reduce false positives due to extraneous factors such as faulty sensors, down gateways or backhaul connectivity observed in real deployment environments. The Mann-Whitney test is complemented with a comparison of quartile coefficient of dispersion and a threshold test of the means before and after the change, to filter out changes due to the above-mentioned factors. Our case studies show that OBCD can significantly reduce false positives by 80% or more compared with the Mann-Whitney test without increasing the detection delay, i.e., the time between event occurrence and its detection.