Bridging the Gap between Smart Home Platforms and Machine Learning using Relational Reference Attribute Grammars

Machine learning improves the development of self-adaptive systems, i.e., to react to unforeseen situations. However, state-of-the-art smart home platforms lack the ability to integrate machine learning models directly into their knowledge base. While there are machine learning approaches for smart home environments, they are usually not integrated nor represented in the middleware of smart environments. Instead, they rely on data obtained previously and use it in a form prepared for a single use case. To remedy this, our aim is to extend existing middleware platforms to integrate machine learning algorithms as first-class concepts of the knowledge base. To achieve this, we employ relational Reference Attribute Grammars to design and implement an integrated runtime model, where machine learning models can be represented and related to elements of the extended knowledge base, e.g., physical entities, location, users, and activities. Consequently, this enables using a state-of-the-art middleware to build a self-adaptive system, which integrates machine learning algorithms enabling both context-awareness and self-awareness. To showcase the feasibility of our approach, we implemented a small smart home scenario using openHAB as a middleware, in which the system learns the preferences of a user using neural networks.