Unobtrusive anomaly detection in presence of elderly in a smart-home environment

Abstract

In the paper we present a method for anomaly detection in user's activities utilizing data from unobtrusive sensors. A service for a smart-home environment using this method adapts to behaviour of a user and may provide alarms to a carer or other responsible person if unusual activity is detected. As an unusual activity we consider: long periods of inactivity, lacking activity, unusual presence and changes in daily activity patterns. Anomaly detection is based on a composition of unsupervised classification technique Self Organizing Maps and next activity prediction employing Markov model. Finally we present a short experimental study realized on a dataset provided by MavHome project.