A Trustworthy Learning Technique for Securing Industrial Internet of Things Systems

Abstract

Since the Industrial Internet of Things (IIoT) networks comprise heterogeneous manufacturing and technological devices and services, discovering advanced cyber threats is an arduous and risk-prone process. Cyber-attack detection techniques have been recently emerged to understand the process of obtaining knowledge about cyber threats to collect evidence. These techniques have broadly employed for identifying malicious events of cyber threats to protect organizations’ assets. The main limitation of these systems is that they are not able to discover and interpret new attack activities. This paper proposes a new adversarial deep learning for discovering adversarial attacks in IIoT networks. Evaluation of correlation reduction has been used as a means of feature selection for reducing the impact of data poisoning attacks on the subsequent deep learning techniques. Feed Forward Deep Neural Networks have been developed using across various parameter permutations, at differing rates of data poisoning, to develop a robust deep learning architecture. The results of the proposed technique have been compared with previously developed deep learning models, proving the increased robustness of the new deep learning architectures across the ToN_IoT datasets.