An Online Model for Detecting Attacks in Wireless Sensor Networks

Abstract

The Internet of Things (IoT) has been employed in a variety of critical fields, including healthcare, geriatric surveillance, self-driving vehicles, and energy management. The most prevalent infrastructure for these applications is Wireless Sensor Networks (WSNs). Although WSNs have intriguing characteristics, the security of such networks is a major concern, especially for applications where confidentiality is extremely crucial. In order to set up WSNs securely, any type of intrusion should be identified before attackers potentially harm the network. However, research findings have demonstrated that existing approaches are ineffective, particularly in detecting attacks in real-time, mostly owing to the accumulation of enormous amounts of data through interconnected devices. Within this interpretation, our intention is to construct a robust Intrusion Detection System (IDS) for analyzing real-time network traffic of WSNs using online learning algorithm while taking into account the network’s resource constraints. To achieve this goal, we investigated the use of an online classifier, namely the Hoeffding Adaptive Tree (HAT), along with the selection of relevant attributes to distinguish four kinds of DoS attacks among normal network traffic: the attacks considered are Blackhole, Grayhole, Flooding, and Scheduling attacks. Among the experimental findings, we determined that utilizing the HAT classifier along with the Chi-squared method of feature selection, detection rate percentages were 86.75%, 80.02%, 94.92%, 99.12% and 99.03% respectively for Flooding, Scheduling, Grayhole, Blackhole and normal case attacks. With an overall accuracy of 99.03%. Based on these findings, it is indeed possible to infer that the HAT classifier is extremely beneficial for categorizing attacks, as it has managed to secure a high detection rate despite the existence of many threats.