ASTRID-Net: SE-enhanced triple attention deep learning framework for IoT and IIoT security

Area of Science:

• Cybersecurity and Network Engineering
• Artificial Intelligence and Machine Learning
• Internet of Things (IoT) and Industrial IoT (IIoT) Security

Background:

• IoT and IIoT networks face significant security vulnerabilities due to their distributed, heterogeneous, and resource-limited nature.
• Cyber attackers exploit these weaknesses for unauthorized access, data leakage, insider misuse, and Distributed Denial of Service (DDoS) attacks.
• Existing security measures struggle to cope with the complexity and scale of modern IoT/IIoT threats.

Purpose of the Study:

• To introduce ASTRID-Net, a novel deep learning architecture for high-accuracy intrusion detection in IoT and IIoT environments.
• To develop a robust framework capable of identifying and mitigating sophisticated cyber threats in real-time.
• To enhance the security and adaptability of cyber-physical systems.

Main Methods:

• ASTRID-Net employs a Triple Attention Hybrid Model integrating multi-scale Convolutional Neural Networks (CNNs) for spatial feature extraction.
• Bidirectional Gated Recurrent Units (BiGRUs) are utilized to capture temporal dependencies within network data.
• A combined channel-temporal attention mechanism prioritizes critical information for improved detection accuracy.

Main Results:

• ASTRID-Net achieved an outstanding accuracy of 99.97% in experimental evaluations.
• Macro-averaged precision, recall, and F1-score exceeded 99.97%, demonstrating superior performance.
• The model significantly outperformed conventional deep learning baselines in intrusion detection tasks.

Conclusions:

• ASTRID-Net proves highly effective and scalable for real-time detection of complex cyber threats in IoT/IIoT infrastructures.
• The model's advanced architecture provides robust security against diverse cyber attacks.
• Findings support the development of more secure and adaptive cyber-physical systems.