Adversarial Evasion Attacks on OCC-Based Machine Learning Intrusion Detection Systems in the Internet of Things

The rapid expansion of Internet of Things (IoT) technologies has transformed interactions between physical and digital systems, driving advancements in smart cities, healthcare, and industrial automation. However, the distributed nature of IoT devices and the vast volumes of data they generate make them prime targets for cyber threats. Intrusion Detection Systems (IDS), enhanced by machine learning, are vital for identifying and mitigating these threats. This paper examines evasion attacks within a one-class classification (OCC) framework, a machine learning technique for anomaly detection, focusing on adversarial attacks like the Fast Gradient Sign Method (FGSM) and Projected Gradient Descent (PGD). The study specifically explores vulnerabilities in OCC models, including autoencoders and support vector machines (SVM), within IoT systems. Experimental results reveal a significant drop in model performance due to adversarial perturbations, highlighting the need for more robust defenses in OCC-based IDS for IoT security.