CIDER: Cyber-Security in Industrial IoT Using Deep Learning and Ring Learning with Errors

Abstract

Traditional security measures such as access control and authentication need to be more effective against ever-evolving threats. Moreover, security concerns increase as more industries shift towards adopting the industrial Internet of things (IIoT). Therefore, this paper proposes secure measures using deep machine learning-based intrusion detection and advanced encryption schemes based on lattice-based cryptography on three-layered cloud-edge-fog IIoT architecture. The novelty of the paper is an integrated security framework for IIoT that combines deep learning-based intrusion detection system (IDS) with lightweight cryptographic protocols. For deep learning, multi-layer perception (MLP), convolutional neural network (CNN), and TabNet were implemented for intruder detection systems from edge to cloud layer, and ring learning with error (RLWE) was proposed for homomorphic encryption to communicate data between fog and edge layer. The evaluation experiments were performed on the Ton_IoT dataset and the results show that the deep learning models have a very good accuracy of around 92% for multiclass attack classification. Moreover, RLWE results show improved encryption time and reduced ciphertext size against standard Learning With Error encryption.