Security Enhancement through Intrusion Detection Systems in Wireless Mesh Networks

Abstract

Any unauthorized action that interferes with the regular functioning of a wired or wireless network is referred to as an intrusion. Wireless mesh networking (WMN) technology has been essential in providing people with ubiquitous access to the Internet at a reasonable and affordable price. Such networks provide universal, high-speed, and cost-effective connection, which is critical for many public services. Malicious attacks, particularly in multi-hop environments, can take advantage of WMNs' decentralized design and accessible media. As a result, it is vital to design these networks with privacy, security, and resiliency. Intrusion detection systems (IDS) are a great way to detect both internal and external attacks. In this work, we used modern technologies such as machine learning to automate IDS. We used publicly available CICIDS2017 data downloaded from the internet for IDS purposes. The raw data was pre-processed to prepare it for analysis. Significant features were chosen using three separate algorithms: Mutual Information (MI), Correlation-based Feature Selection (CFS), and Particle Swarm Optimization (PSO). These selected features were then fed into Machine Learning (ML) models, specifically Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Random Forest (RF), and Attention-based AutoEncoder (AAE). Accurate IDS relies on determining the best combination of Feature Selection (FS) and ML model. Through experimental study, we discovered that the combination of PSO and AAE produced the best accuracy of 99.54%, with the lowest false prediction rates of 0.48% and 0.44%. Following closely, the combination of PSO and RF yielded higher results, with an accuracy of 98.9%.