Multi-inhabitant and eXplainable Activity Recognition in Smart Homes

The sensor-based detection of Activities of Daily Living (ADLs) in smart home environments can be exploited to provide healthcare applications, like remotely monitoring fragile subjects living in their habitations. However, ADLs recognition methods have been mainly investigated with a focus on single-inhabitant scenarios. The major problem in multi-inhabitant settings is data association: assigning to each resident the environmental sensors' events that he/she triggered. Furthermore, Deep Learning (DL) solutions have been recently explored for ADLs recognition, with promising results. Nevertheless, the main drawbacks of these methods are their need for large amounts of training data, and their lack of interpretability. This paper summarizes some contributions of my Ph.D. research, in which we are designing explainable multi-inhabitant approaches for ADLs recognition. We have already investigated a hybrid knowledge-and data-driven solution that exploits the high-level context of each resident to perform data association. Currently, we are studying semi-supervised techniques to mitigate the data scarcity issue, and eXplainable Artificial Intelligence (XAI) methods to make DL classifiers for ADLs more transparent.