An online pattern based activity discovery: In context of geriatric care

Abstract

The behavioral analysis of an elderly person living independently is one of the major components of geriatric care. The day-long activity monitoring is a pre-requisite of the said analysis. Activity monitoring could be done remotely through the analysis of the sensory data where the sensors are placed in strategic locations within the residence. Most of the existing works use supervised learning. But it becomes infeasible to prepare the training dataset through repeated execution of a set of activities for a geriatric person. Moreover, the geriatric people are annoyed to use wearable sensors. Thus it becomes a challenge to discover the activities based on only ambient sensors using unsupervised learning. Pattern-based activity discovery is a well-known technique in this domain. Most of the existing pattern based methods are offline as the entire data set needs to be mined to find out the existing patterns. Each identified pattern could be an activity. There are a few online alternatives but those are highly dependent on prior domain knowledge. In this paper, the intention is to offer an online pattern based activity discovery that performs satisfactorily without any prior domain knowledge. The exhaustive experiment has been done on benchmark data sets ARUBA, KYOTO, TULUM and the performance metrics ensure the strength of the proposed technique.