Deep Kronecker hashing generative adversarial network for intrusion detection in IoT against adversarial attacks

The rapid growth of wireless networks accelerates the adoption of the Internet of Things (IoT) devices in applications, such as industrial automation, healthcare, and smart homes, ensuring seamless connectivity and real-world data transmission. However, the widespread adoption of interconnected IoT systems makes them main targets for cyber threats, particularly adversarial attacks, which exploit system vulnerabilities by manipulating data and compromising privacy and security. Therefore, intrusion detection systems (IDSs) play a vital role in monitoring IoT network behavior and maintaining data leakage, malicious attacks, and unauthorized access. Owing to the heterogeneous nature and resource-constrained characteristics of IoT devices, intrusion detection against adversarial attacks a challenging task. To tackle these challenges, this article develops a novel Deep Kronecker Hashing Generative Adversarial Network (DKHashGAN)-based intrusion detection against adversarial attacks in IoT scenarios. Initially, a Kitsune-based Network Intrusion Detection System (NIDS) is simulated to generate network traffic data. The input log file is normalized by Comprehensive normalization, followed by feature selection based on the Chord distance. Finally, intrusion detection is conducted using the developed DKHashGAN model, which combines a Deep Kronecker Network (DKN) with a Hashing Generative Adversarial Network (HashGAN) to improve detection capability. The robustness of DKHashGAN is evaluated against Projected Gradient Descent (PGD), Basic Iterative Method (BIM), and Fast Gradient Sign Method (FGSM) attacks using K-fold and the percentage of perturbation. Experimental outcomes indicate that the developed model achieves the optimal accuracy, True Positive Rate (TPR), True Negative Rate (TNR), and False Positive Rate (FPR) of 96.43%, 95.90%, 97.91%, and 2.08%, for the FGSM attack using K-fold of 9.