Sketch Based Anomaly Detection Scheme in Wireless Sensor Networks

Abstract

The constrained capacity of wireless sensor nodes and harsh, unattended deploy environments make the data collected by sensor nodes usually unreliable. In this paper, we propose a sketch based anomaly detection scheme in order to detect the anomaly data values. It first partitions the whole sensor network into several clusters in which the cluster members are physically adjacent and data correlated. Then, the cluster header collects the count-min sketch of each cluster member and compares it with its own sketch in the form of kullback-leibler divergence. The experiment shows that the proposed anomaly detection scheme can provide a high detection accuracy ratio and a low false alarm ratio.