Model-Based Approach for Anomaly Detection in Smart Home Inhabitant Daily Life

Due to population ageing, the number of people requiring monitoring and specific health care increased. Nevertheless, due to social or financial reasons, most of this population prefers to stay at home. Thanks to recent improvement in Ambient Assisted Living (AAL) and smart home technologies, new opportunities emerge to develop Health at Home (HaH) solutions. One of the main objectives of HaH is to offer to smart home inhabitants a life and health quality similar to patients in specialized medical institutions. To achieve this task, it is proposed in this paper an approach to monitor frail people at home to detect behavioral changes which can be symptoms of health problems. The method is model-based, working with Stochastic Timed Automata as it is an efficient tool to model activity duration and ordering habits, which are key features of human behavior. Once a model depicting life habits is built, it is proposed to use normalized likelihood and probabilistic distribution to evaluate the consistency between the regular life habits and newly observed behavior. Appearance of a inconsistent behavior may testify the evolution or apparition of a medical symptom. A case study demonstrates the relevancy of the proposed approach, and several scenarios highlight the possibility to have clear enough information on inhabitant behavior to identify a limited number of responsible medical disorder.