Semantic Modelling of Multivariate Time-Series Data in Cognitive IoT

With the use of sensors and intelligent devices to assist interactions with the physical environment, we observe a significant rise in the amount of sensory input, which is continual, volatile, sometimes actual, and diverse in structure. The Internet of Things (IoT) encapsulates this phenomenon. In addition, when IoT integrates cognition into system design, it is called Cognitive IoT (CIoT). CIoT requires techniques to autonomously extract semantic patterns from plain sensory data for efficient service delivery. Only a little scientific attention has been paid to automatically identifying semantic patterns from a massive time-series data set. This research presents a scaffold for real-time automatic semantic pattern extraction in the Cognitive Internet of Things. It utilizes the Symbolic Aggregate Approximation (SAX) technique for sensory data with a multivariate time series. Further, the output of SAX is used to create a symbol table with their frequency. This frequency table is transformed into its corresponding probabilistic value. Finally, this value is used to generate the semantic pattern, which can be further processed using a rule-based method to create a machine-interpretable pattern. Experimental evaluation of this research study reveals that the proposed method can extract semantic patterns from heterogeneous multivariate sensory data.