SECURING INDUSTRIAL IOT: GCN-BASED IDS IMPLEMENTATION AND A REVIEW OF TESTING FRAMEWORKS

Abstract

Cyber-attacks on Industrial IoT systems can result in severe consequences such as production loss, equipment damage, and even human casualties and hence security is of utmost concern in this application of IoT. This thesis, presents an approach for network security, intrusion detection that utilizes the spatial attributes of a network in attempt overcome the limitations discovered through literature review of various studies in Intrusion Detection and testing frameworks. For this graph-based neural network have been used that was seen promising in modelling complex relationships between graphical entities, making them a suitable approach for IDS in interconnected systems. Our approach leverages a graph representation of network traffic, that is used as an input for neural network through the use of convolution operation. Our approach makes use of flow features of the network in relation with the neighbouring flows in contrast to other machine learning models that uses flow features independent to each other. This work has been evaluated primarily on Edge-IIoT 2022, dataset and compared with existing well-known datasets and machine learning methods. The results show that our approach achieved average 5.49% improved F1-score, compared with other standard existing methods with our model having highest F1-Score of 0.996. Further research and development in this area will advance the field of IIoT security and enhance the resilience of industrial systems in the face of evolving threats.