Activity Discovery and Detection of Behavioral Deviations of an Inhabitant From Binary Sensors

The aim of this paper is to improve the autonomy of medically monitored patients in a smart home instrumented only with binary sensors; overwatching the disease evolution, that can be characterized by behavior changes, is helped by detecting the activities the inhabitant performs. Two contributions are presented. On one hand, using sequence mining methods in the flow of sensor events, the most frequent patterns mirroring activities of the inhabitant are discovered; these activities are then modeled by an extended finite automaton, which can then be used for activity recognition and generate activity events. On the other hand, given the set of activities that can be recognized, another automaton is built to model requirements from the medical staff supervising the inhabitant; it accepts activity events, and residuals are defined to detect any behavior deviation. The whole method is applied to the dataset of Domus, an instrumented smart home.