Composite event detection and identification for WSNs using General Hebbian Algorithm

In this paper, we propose an on-line technique for in-network, distributed and composite event detection and identification for streaming sensor data in resource constrained Wireless Sensor Networks (WSNs). We use General Hebbian Algorithm (GHA) to find out principal components of a multi-attribute input data which has a linear complexity as opposed to quadratic complexity with eigen value decomposition (EVD). This allows for on-line computation of percentage contributions of individual attributes towards detected event. Comparison with other event detection techniques shows that our scheme incurs low communication overhead as compared to some state-of-the-art schemes. Moreover, our hyper-ellipsoidal clustering based event detection algorithm is shown to achieve high detection rates (DRs) of over 98.88% and very low false positive rates (FPRs) of below 0.01%. Our simulation results and the hardware implementation also show that the accuracy of proposed identification scheme is in strong agreement with EVD based techniques, proving it to be a successful event identification method for WSNs.