A Hybrid CNN-LSTM Model for IIoT Edge Privacy-Aware Intrusion Detection
Erik Miguel de Elias, Vinícius Sanches Carriel, Guilherme Werneck de Oliveira, Aldri Santos, Michele Nogueira, Roberto Hirata, Daniel Macêdo Batista
Abstract
Security is a critical issue in the context of IoT and, more recently, of Industrial IoT(IIoT) environments. To mitigate security threats, Intrusion Detection Systems have been proposed. Still, most of them can achieve high accuracy only by having access to the application layer of the flows, which is problematic in terms of privacy. This paper presents a neural network model based on a hybrid CNN-LSTM architecture to detect several attacks in the network traffic at the Edge of IIoT using only features from the transport and network layers. Besides improving privacy, the proposal achieves 97.85% average accuracy when classifying the traffic as benign or malicious and 97.14% average accuracy when classifying 15 specific attacks in a dataset containing IIoT traffic. Moreover, all the code produced is available as free software, facilitating new studies and the reproduction of the experiments.