A Review of Intrusion Detection for Internet of Things Using Machine Learning
Ekwueme Chika Paul, Amrita Amrita · 2024
As Internet of Things (IoT) networks proliferate across consumer and industrial sectors, safeguarding these interconnected cyber-physical systems is imperative. IoT ecosystems present abundant targets for malicious intrusions including denial-of-service, man-in-the-middle, crypto and access attacks due to their scale and dynamic topology. Such intrusions can severely disrupt sensing, monitoring and control functions, enabling catastrophic cascading failures. Consequently, IoT platforms require intelligent intrusion detection system (IDS) to identify anomalous behaviors and known attack patterns. Conventional signature-based detection relying on expert rules proves inadequate against evolving attacks. In contrast, data-driven machine learning (ML) algorithms offer versatile capabilities for securing IoT by modeling normal vs anomalous traffic and information flows based on supervised, semi-supervised or unsupervised learning. Critical analysis reveals ML-driven intrusion detection frameworks achieve high accuracy in controlled settings. However, challenges related to computational burdens, adversarial attacks, concept drift and lack of model standardization need resolution before widespread field deployment. As IoT ecosystems continue proliferating amidst escalating threats, our systematic review helps identify limitations and promising directions for future research on ML-powered IDSs. This paper presents a comprehensive analysis of adopting ML techniques like neural networks, random forest, support vector machines, ensemble models and deep learning architectures to engineer anomaly detection and misuse detection modules tailored for IoT environments.