IoT-based human action prediction and support

Abstract

It is an important topic in Active and Assisted Living (AAL) research and development to support elderly people suffering from memory impairment in their daily activities. A promising approach to such support is providing memory aids based on knowledge of how the person to be supported usually (i.e., in an unimpaired condition) copes with her/his daily activities. Such knowledge may be captured by IoT solutions,appropriately structured and stored in a knowledge base, and exploited when the need of support is detected. Determining the best help for a given situation implies decision-making, since the actions– flow (behavior) of an activity usually involves probabilistic branches: An automated system needs to decide which of the possible next actions is best suited for the user in a given situation. Problems of this nature involve uncertainty levels that have to be dealt with. Many approaches to this problem exploit statistical data only, thus ignoring important semantic data as, for instance, are provided by Ontologies. However, ontologies do not support reasoning over uncertainty natively. In this paper, we present a probabilistic semantic model that represent information from IoT sources and enables reasoning over uncertainty without losing semantic information. This model is implemented as an extension of the Human Behavior Monitoring and Support (HBMS) approach that provides a conceptual ”human cognitive model” for representing the user–s behavior and its context in her/his living environment. The performance of this approach was evaluated using real data collected from a smart home prototype equipped with installable sensors and IoT devices. The experiments provided promising results which we will discuss regarding limits and challenges to overcome.