A Dynamic Graph-based Cluster Ensemble Approach to Detect Security Attacks in Surveillance Network

Abstract

Wireless sensor networks (WSNs) are underlying network infrastructure for a variety of mission-critical surveillance applications. The network should be tolerant of unexpected failures of sensor nodes to meet the Quality of Service (QoS) requirements of these applications. One major cause of failure is active security attacks such as Denial of Service (DoS) attacks. This paper models the problem of detecting such attacks as an anomaly detection problem in a dynamic graph. The problem is addressed by employing a voting based cluster ensemble approach called the K-Means Spectral and Hierarchical ensemble (KSH) approach. The experimental result shows that KSH detected DoS attacks with better accuracy when compared to baseline approaches. sectionIntroduction and Motivation WSNs play a vital role in a variety of mission-critical surveillance applications, such as military surveillance. These applications demand different QoS, such as energy efficiency, coverage, and connectivity from the underlying network. To meet these QoS requirements, WSNs should be tolerant of sensor node failures. Active security attacks such as DoS attacks are one major cause of such failures. The famous Maroochy water treatment and Ukrainian power grid attacks are good instances of active security attacks over wireless sensor networks. Active security attacks are more dangerous in terms of severity it creates in the network. For instance, such an attack on WSNs deployed for military surveillance applications can lead to physical intrusions to happen without being undetected. WSNs are prone to such attacks due to its inherent constraints such as limited bandwidth, lack of tamper-proof hardware, and lack of physical line of defense such as Firewalls. Cryptographic solutions are one commonly used method in the literature to address these attacks. But, such solutions are not a viable option to detect attacks in resource constraint WSNs. A lightweight and energy-efficient intruNetwork modelling phase