A Secure Adversarial Attack Detector Framework for Monitoring Network Intrusion in IoT Environment

As computer networks expand, the demand for robust security measures intensifies due to increasing threats that jeopardize network integrity and confidentiality. This is particularly critical in the context of the Internet of Things (IoT), where interconnected devices significantly broaden the attack surface. Despite the growing adoption of Intrusion Detection Systems (IDS) enhanced with machine learning (ML), these systems face a critical vulnerability: they can be easily deceived by adversarial attacks specifically crafted to evade detection. This paper addresses the problem of adversarial robustness in ML-based IDS, particularly under white-box attack scenarios, by introducing the Adversarial Attack Detector (AAD) framework. AAD incorporates a multi-layered defense mechanism, including a request validator, a machine learning-based Adversarial Discriminator (AD), and an Enhanced Intrusion Detection System (EIDS). The EIDS combines adversarially trained Multi-Layer Perceptrons, Convolutional Neural Networks, and Long Short-Term Memory networks to improve detection accuracy and robustness. The scope of this study is centered on securing IoT environments, where conventional IDS models struggle due to the limitations of constrained devices and heterogeneous data. Our empirical analysis shows that the AAD framework significantly mitigates the impact of various white-box adversarial attack methods, and outperforms traditional and single-model approaches. AAD’s components, particularly the EIDS, significantly recover model performance under adversarial settings, improving accuracy from near-failure levels to near-perfect classification in some scenarios.