A cognitive data stream mining technique for context-aware IoT systems
Dinithi Nallaperuma, Daswin De Silva, Damminda Alahakoon, Xinghuo Yu · IECON 2017 - 43rd Annual Conference of the IEEE Industrial Electronics Society · 2017
IoT systems deployed in industrial and smart factory settings generate large volumes of data at high velocity. Context awareness is mandatory for knowledge discovery and actionable insights from such high-velocity, high-volume IoT data streams. Changes to the context of a data stream are represented in the underlying data distribution. Research in concept drift aims to detect and adapt to such changes in a data distribution. Concept drift detection can be extended to suit ad hoc Big Data streams generated by IoT systems, by introducing the cognitive principles of learning. This paper proposes an unsupervised incremental learning algorithm for detection and adaption of concept drift based on the cognitive principles of learning. It executes in automated time windows, detects concept drift using movement in space and determines the type of concept drift using movement in time. The algorithm was applied to a Big Data set representing an IoT system for urban vehicular movement and traffic. Results confirm that the proposed algorithm generates context-awareness by detection and adaptation to concept drift in high-volume, high-velocity IoT systems.