Improved Intrusion Detection Based on Hybrid Deep Learning Models and Federated Learning.

Area of Science:

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

Background:

• Network Intrusion Detection Systems (NIDS) are crucial for Industrial Internet of Things (IIoT) security.
• Deep learning models for NIDS require large datasets, which are often limited locally and raise privacy concerns when centralized.

Purpose of the Study:

• To develop a novel federated learning (FL) based approach for enhancing network intrusion detection accuracy in IIoT environments.
• To ensure robust data privacy protection during the training of intrusion detection models.

Main Methods:

• A deep learning intrusion detection model combining convolutional neural networks (CNNs) with attention mechanisms was developed for IIoT.
• Variational autoencoders (VAEs) were integrated for enhanced data privacy.
• A federated learning (FL) framework facilitated collaborative training of a shared model across multiple IIoT clients without raw data sharing.

Main Results:

• The proposed FL approach significantly improved detection accuracy, precision, and reduced the false-positive rate (FPR).
• Experimental validation on a real-world IoT network intrusion dataset confirmed the model's effectiveness.
• The method demonstrated superior performance compared to traditional local training and existing NIDS models.

Conclusions:

• The novel FL-based deep learning model effectively addresses the challenges of limited data and privacy concerns in IIoT network intrusion detection.
• This approach offers a promising solution for securing IIoT systems by improving NIDS performance while safeguarding sensitive data.