A Lightweight Multilayer Machine Learning Detection System for Cyber-attacks in WSN
Shereen Ismail, Diana W. Dawoud, Hassan Reza
Abstract
A key technology that empowers low-cost and low-power Internet of Things (IoT) systems implementation is the Wireless Sensor Network (WSN); however, WSNs are susceptible to several types of cyber-attacks, leading to new security risks that require proper detection and mitigation techniques design. Machine learning is one of the most promising solutions for developing effective detection systems. This paper presents a lightweight multi-layer machine learning detection system to mitigate cyber-attacks that target WSNs. We aim to address internal WSN attacks with the assistance of a mobile robot. The multi-layer detection system consists of two machine learning models deployed at monitor nodes and Base Station (BS). We used a Naive Bayes algorithm as First-layer detection for binary classification, and a LightGBM algorithm as Second-layer detection for multi-class classification. The proposed system can detect four network-layer internal Denial-of-Service (DoS) attacks observed in WSN-DS dataset. The monitor nodes detect the attack when it occurs, and the mobile robot moves to the cluster where the attack is detected so it can route the updates to the BS for Second-layer detection and further investigation.