Focal Causal Temporal Convolutional Neural Networks: Advancing IIoT Security with Efficient Detection of Rare Cyber-Attacks

Abstract

The Industrial Internet of Things (IIoT) deals with vast amounts of data that must be safeguarded against tampering or theft. Identifying rare attacks and addressing data imbalances pose significant challenges in the detection of IIoT cyberattacks. Innovative detection methods are important for effective cybersecurity threat mitigation. While many studies employ resampling methods to tackle these issues, they often face drawbacks such as the use of artificially generated data and increased data volume, which limit their effectiveness. In this paper, we introduce a cutting-edge deep binary neural network known as the focal causal temporal convolutional neural network to address imbalanced data when detecting rare attacks in IIoT. The model addresses imbalanced data challenges by transforming the attack detection into a binary classification task, giving priority to minority attacks through a descending order strategy in the tree-like structure. This approach substantially reduces computational complexity, surpassing existing methods in managing imbalanced data challenges in rare attack detection for IoT security. Evaluation of various datasets, including UNSW-NB15, CICIDS-2017, BoT-IoT, NBaIoT-2018, and TON-IIOT, reveals an accuracy of over 99%, demonstrating the effectiveness of FCTCNNs in detecting attacks and handling imbalanced IoT data with efficiency.

Keywords

• IoT security
• focal causal temporal CNNs (FCTCNNs)
• imbalanced data problem
• deep learning
• intrusion detection system (IDS)