Related Experiment Videos
Lightweight and Energy-Aware Intrusion Detection for Industrial IoT Using TinyML and Edge AI.
Laila Nassef1, Mohammed Ibrahim Alghamdi2, Slim Ben Chaabane3
1Department of Computer Science,Faculty of Computing and Information Technology, King Abdulaziz University, Jeddah 21589, Saudi Arabia. Lmohamed@kau.edu.sa.
Scientific Reports
|June 15, 2026
Summary
This study introduces a hybrid Intrusion Detection System (IDS) using Graph Attention Networks (GAT) and Bidirectional Gated Recurrent Units (BiGRU) for enhanced Industrial Internet of Things (IIoT) security. The framework achieves high accuracy in detecting diverse cyber threats while preserving data privacy through Federated Learning (FL).
Area of Science:
- Cybersecurity
- Machine Learning
- Network Security
Background:
- Industrial Internet of Things (IIoT) ecosystems are rapidly expanding, necessitating robust security measures.
- Critical infrastructures face evolving cyber threats, demanding scalable and reliable Intrusion Detection Systems (IDS).
- Existing IDS solutions often struggle with privacy concerns and decentralized deployment in complex IIoT environments.
Purpose of the Study:
- To propose a novel hybrid IDS framework for privacy-preserving distributed detection in IIoT.
- To enhance the accuracy and scalability of IDS by combining Graph Attention Networks (GAT) and Bidirectional Gated Recurrent Units (BiGRU).
- To optimize the framework using Grey Wolf Optimizer (GWO) and Federated Learning (FL) for improved performance and data privacy.
Main Methods:
- A hybrid IDS framework integrating GAT for structural analysis and BiGRU for temporal pattern recognition.
- Optimization using Grey Wolf Optimizer (GWO) for automated hyperparameter tuning and faster convergence.
- Enhancement through Federated Learning (FL) for privacy-preserving, decentralized model training on distributed IIoT devices.
- Integration of an Explainable AI (XAI) module leveraging GAT's attention mechanism for improved interpretability.
Main Results:
- Achieved detection accuracies of up to 95% across various attack scenarios (DDoS, APTs, Zero-Day exploits).
- Demonstrated improved scalability and reduced communication overhead (20% lower in a 10-node simulation) via FL.
- Attained competitive offline performance with F1-scores up to 0.94 on benchmark datasets (Edge-IIoTset, CICIoT2023, RT-IoT2022).
- Identified inference latency on resource-constrained edge hardware (120-180 ms/sample on Raspberry Pi 4) as a limitation for strict real-time applications.
Conclusions:
- The proposed hybrid IDS framework offers a promising solution for scalable, privacy-preserving threat detection in IIoT environments.
- The combination of GAT, BiGRU, GWO, and FL effectively addresses accuracy, scalability, and privacy challenges.
- Further research is needed in model compression and real-world validation to overcome edge hardware latency limitations for mission-critical applications.