Analyzing Adversarial Attacks Against Deep Learning for Intrusion\n Detection in IoT Networks

Adversarial attacks have been widely studied in the field of computer vision\nbut their impact on network security applications remains an area of open\nresearch. As IoT, 5G and AI continue to converge to realize the promise of the\nfourth industrial revolution (Industry 4.0), security incidents and events on\nIoT networks have increased. Deep learning techniques are being applied to\ndetect and mitigate many of such security threats against IoT networks.\nFeedforward Neural Networks (FNN) have been widely used for classifying\nintrusion attacks in IoT networks. In this paper, we consider a variant of the\nFNN known as the Self-normalizing Neural Network (SNN) and compare its\nperformance with the FNN for classifying intrusion attacks in an IoT network.\nOur analysis is performed using the BoT-IoT dataset from the Cyber Range Lab of\nthe center of UNSW Canberra Cyber. In our experimental results, the FNN\noutperforms the SNN for intrusion detection in IoT networks based on multiple\nperformance metrics such as accuracy, precision, and recall as well as\nmulti-classification metrics such as Cohen's Kappa score. However, when tested\nfor adversarial robustness, the SNN demonstrates better resilience against the\nadversarial samples from the IoT dataset, presenting a promising future in the\nquest for safer and more secure deep learning in IoT networks.\n