Distributed Event Identification for WSNs in Non-Stationary Environments

This paper proposes a novel scheme to estimate the percentage contribution of different attributes in a detected event (a process termed as event identification) for streaming multi-attribute data in WSNs. The proposed event detection and identification algorithm takes into account correlation among sensed attributes as well as the spatio-temporal correlations with similar attributes measured by neighboring nodes. Moreover we update our statistical parameters in an iterative manner such that the dynamics of non- stationary environments are taken into account. We test our leave one out (LOO) event identification approach with simulations on both synthetic and real data sets and an implementation on off-the- shelf WizziMotes. The experimental results show that our detection scheme outperforms state of the art schemes by showing detection rates (DRs) of more than 98\% and false positive rates (FPRs) of less that 2\%. Moreover, our event identification approach effectively determines the contribution of both correlated and uncorrelated attributes in an event of interest. The identification has also been shown to be in strong agreement with previous computationally complex benchmark PCA based event identification approaches.