NIDS-FGPA: A federated learning network intrusion detection algorithm based on secure aggregation of gradient...

Area of Science:

• Cybersecurity
• Network Security
• Industrial Internet of Things (IIoT)

Background:

• IIoT security is critical due to increasing network threats.
• Existing intrusion detection systems struggle with incomplete data, missing labels, and high communication overhead.
• Need for robust and secure intrusion detection in IIoT environments.

Purpose of the Study:

• To propose a federated learning-based intrusion detection algorithm (NIDS-FGPA) for IIoT security.
• To address challenges of incomplete data, missing labels, and communication overhead.
• To enhance the security and efficiency of IIoT networks.

Main Methods:

• Federated learning architecture combined with Paillier homomorphic encryption for secure training.
• Gradient Similarity Model Aggregation (GSA) to reduce communication overhead by dynamic model update selection and weighting.
• A deep learning model utilizing 2D convolutional neural networks and bidirectional gated recurrent units (2DCNN-BIGRU) to process complex network traffic data.

Main Results:

• Achieved 94.5% accuracy on the Edge-IIoTset dataset.
• Achieved 99.2% accuracy on the CIC IoT 2023 dataset.
• Demonstrated ability to identify and capture complex network attacks.

Conclusions:

• The NIDS-FGPA model effectively enhances IIoT network security.
• The proposed methods successfully address limitations of existing intrusion detection systems.
• Federated learning with homomorphic encryption and GSA offers a promising solution for secure and efficient IIoT intrusion detection.