Stream Processing With Concept Drift for Event Identification in Sensors Enabled IoT Environment

A streaming IoT environment like health monitoring system is characterized by continuous generation of data at a high rate. This stream of data can neither be stored on the nodes nor transmitted to the sink due to resource constraints. This demands continuous perusal of the data stream at nodes for in-situ event detection. In this WSN scenario, existing schemes that rely on inference based on stored data on the nodes or the sink become in-efficacious. In this paper, temporal and spatial correlation in such data streams is exploited to effectively reduce the use of these resources in decision making process. We propose an All In One Stream Processing (AIOSP) scheme which employs sampling using temporal correlation, load shedding using spatial correlation and approximation using drift and diffusion for efficient decision making process in data streaming environment where a large amount of data is continually generated by the sensor nodes. The performance of proposed scheme is evaluated through simulations over varied sample size and is found to ensure 85% to 95% accuracy with small sample size.