Anomaly Detection using Machine Learning Techniques in Wireless Sensor Networks

Abstract

The emplacement of wireless sensor networks (WSNs) has increased dramatically in the last few years. WSNs attract many enterprises to use them in various applications due to their compact size and low cost. Environmental monitoring, building security, and precision agriculture are just a few examples among many other industries. Since most of them are located in hostile and unmanned environments, WSNs pose significant security risks. Many options have been proposed to protect the privacy of data during its transport from sensors to the base station in order to enable secure data processing in WSNs. The aim of this work is attack detection, a key responsibility for network and data protection. To keep wireless sensor networks secure and free from malicious attacks, anomaly detection is a crucial task. Currently, researchers use various machine learning techniques to identify anomalies using offline learning algorithms. However, online learning classifiers have not received enough attention in the literature. Our goal is to offer an intrusion detection model that works with the unique properties of WSNs. This approach is based on online passive aggressive classifier and information gain ratio. First, the relevant aspects of the sensor data are selected using the information gain ratio. The Online Passive Aggressive algorithm also learns to recognize and categorize different types of Denial of Service attacks. The test was performed using a wireless dataset from a wireless sensor network detection system (WSN-DS) that was used for the investigation. The proposed ID-GOPA model achieves a 96% detection rate in deciding whether a network is operating normally or vulnerable to attacks of any kind. In addition to 99% for normal operation, the detection accuracy for planning, gray hole attacks, flooding, and black hole attacks is 86%, 68%, 63%, and 46%, respectively. These findings show that under certain circumstances, our methodology based on offline learning can replace online learning and offer efficient WSN anomaly detection.