Data-driven generation of rule-based behavior models for an Ambient assisted living system

In this paper we introduce an approach for modeling the typical behavior of inhabitants in smart homes. The presented modeling process is data-driven and based on unsupervised learning methods. The models consist of association rules that are automatically generated from collected sensor telegrams by data mining. The intended application for such models is the detection of alterations in the mid- or long-term behavior indicating possible changes in health conditions of users of Ambient Assisted Living systems. We successfully applied the modeling approach to real world sensor data recorded in permanently inhabited flats.