Continuous, Online Anomaly Region Detection and Tracking in Networks.

Shuiyuan Xie,Xiuli Ma,Shiwei Tang
DOI: https://doi.org/10.1007/978-3-642-39527-7_9
2013-01-01
Abstract:In many real networks, the detection and tracking of unusual phenomena, such as the diffusion of contamination and the spreading of disease, is one of the key feature users are great interested in, which is called anomaly with technical terms. In this paper, we present a framework to detect and track anomaly region continuously. First, we build a state transition graph to summarize network's operating regularity, that is, network stays in a state for a period of time and alternates among states over and over again, which exists in many real networks. Second, we employ the state transition graph to predict network's next state. While comparing expected state and current state, we present suspicious region and its anomaly probability. We evaluate our approach on a real water distribution network from the Battle of the Water Sensor Network (BWSN). Experiments show that our approach is effective, efficient and scalable to detect and track anomaly region. © 2013 Springer-Verlag.
What problem does this paper attempt to address?