Focal Causal Temporal Convolutional Neural Networks: Advancing IIoT Security with Efficient Detection of Rare...

Area of Science:

• Cybersecurity
• Machine Learning
• Industrial Internet of Things (IIoT)

Background:

• The Industrial Internet of Things (IIoT) generates vast data requiring robust security against tampering and theft.
• Detecting rare cyberattacks in IIoT is challenging due to imbalanced datasets, often inadequately addressed by traditional resampling methods.
• Existing methods face limitations including artificial data generation and increased computational load.

Purpose of the Study:

• To introduce an innovative deep binary neural network for detecting rare cyberattacks in IIoT environments.
• To address the significant challenge of imbalanced data in IIoT cybersecurity threat detection.
• To improve the efficiency and effectiveness of IIoT security systems against sophisticated threats.

Main Methods:

• Development of a focal causal temporal convolutional neural network (FCTCNN), a deep binary neural network.
• Transformation of attack detection into a binary classification task, prioritizing minority (rare) attacks.
• Implementation of a descending order strategy within a tree-like structure to manage imbalanced data and reduce computational complexity.

Main Results:

• The FCTCNN model demonstrated superior performance in handling imbalanced data for rare attack detection.
• Achieved an accuracy exceeding 99% across multiple benchmark datasets including UNSW-NB15, CICIDS-2017, BoT-IoT, NBaIoT-2018, and TON-IIOT.
• Significantly reduced computational complexity compared to existing methods for IIoT security.

Conclusions:

• The FCTCNN is a highly effective and efficient solution for detecting rare cyberattacks in imbalanced IIoT datasets.
• The proposed model offers a substantial advancement in IIoT cybersecurity, enhancing threat mitigation capabilities.
• This approach provides a scalable and accurate method for securing IIoT systems against evolving cyber threats.