Transferable intrusion detection model for industrial Internet based on deep learning

The diverse characteristics of the Industrial Internet of Things (IIoT) have significantly influenced the evolution of Industrial Internet Intrusion Detection Systems (IIDS). Current IIDS solutions are unable to effectively migrate to relevant domains and identify zero-day vulnerabilities due to a lack of comprehensive, high-dimensional characterization of the attack types exposed in the network and models built on outdated data sets. In this paper, we introduce a new transferable IIDS model with deep learning at its core, which combines efficient feature mapping technology with cascade models and transfer learning (TL) to solve the above problems. Our approach combines Autoencoders (AE) to identify suitable features and (CFBPNN) for attack identification and classification detection. We conduct a set of experiments on five popular IoT and IIoT datasets: Edge-IIoTSet, NSLKDD+, UNSW-NB15, WUSTL-IIoT-2021 and X-IIoTID. We calculated the Accuracy, Recall, Precision, F1-source, MCC and AUC of this method model. The results show that our method provides over 97% accuracy and 95% MCC. TL enhances the ease of use of our model. When tested on the NSLKDD+ dataset, the pre-trained model on the Edge-IIoTSet dataset resulted in an increase in MMC from -10.69% to 93.01%, and other values were greatly improved.