A machine learning scheme to assess the radio propagation conditions on wireless communication networks

Abstract Wireless communication has become the primary means of electronic communication around the world. Particularly, we focus on the radio propagation conditions of a wireless sensor network (WSN), our case study, where common methods exist for measurements of WSN’s performance. However, these methods are expensive and not fully automated, and therefore they have severe drawbacks and limitations. Thus, this paper describes an automatic, simple, and fast-answer system for solving practical radio-communication problems by resorting to two branches of artificial intelligence: machine learning and knowledge-based systems (KBS). The non-incremental unsupervised learning module classifies a dataset into clusters that are then computed in order to induce concepts, which are specific features shared by every dataset within a cluster. These concepts serve to construct implication rules being the knowledge used by the KBS in charge of answering queries about WSN in radio propagation conditions. Furthermore, given that WSN is a dynamic environment, our system also incorporates an incremental supervised learning module in order to process a new received reduced dataset, and potentially to induce new concepts and update the knowledge base of implication rules. Such process is performed unchanging the inference engine module of our KBS. Finally, we provide illustrating examples of our approach’s abilities for query-answering, highlighting that its response time fulfills restricted real-world conditions thanks to the conception of polynomial-complexity algorithms on which our system runs, and which are among our contributions .