Related Experiment Video
Updated: Jul 11, 2026
05:47
Evidence-based Knowledge Synthesis and Hypothesis Validation: Navigating Biomedical Knowledge Bases via Explainable AI and Agentic Systems
Published on: June 13, 2025
An optimized graph neural network approach for robust and explainable IoT intrusion detection against adversarial
Uzma Ghulam Mohammad1, Adil Afzal2, Saleh Alghamdi3
1Department of Software Engineering, Lahore Garrison University, Lahore, 54792, Pakistan.
Scientific Reports
|May 11, 2026
Summary
This study introduces a novel intrusion detection system using Graph Neural Networks and adversarial training to combat adaptive cyberattacks. The enhanced framework achieves up to 97% accuracy, significantly outperforming existing methods in detecting sophisticated threats.
Area of Science:
- Cybersecurity
- Machine Learning
- Network Intrusion Detection
Background:
- Machine learning-based security systems are vulnerable to adaptive adversarial attacks.
- Existing DDoS detection methods lack resilience against adversarial manipulation.
- There is a critical need for robust and reliable defense mechanisms against evolving cyber threats.
Purpose of the Study:
- To present a reliable and comprehensible intrusion detection paradigm against adversarial attacks.
- To improve the transparency and reliability of intrusion detection systems.
- To enhance the robustness of IoT intrusion detection against complex traffic connections.
Main Methods:
- Utilized Graph Neural Networks (GNNs), Deep Neural Networks (DNN), DeepFool, and First Gradient Sign Method (FGSM) for adversarial training.
- Introduced a novel adversarial dataset, AdvCICDDoS2019, with four types of adversarial attacks.
- Employed SHAP and LIME for interpretability and combined DeepFool with FGSM for enhanced robustness.
Main Results:
- The proposed framework exceeded current methods by 4% to 12% across various attack scenarios.
- Achieved a detection accuracy of up to 97% under hostile settings, demonstrating resilience against sophisticated traffic.
- Explainable adversarial defense mechanisms and graph-aware learning significantly improved model reliability and transparency.
Conclusions:
- The developed intrusion detection system offers a more robust and reliable defense against adaptive adversarial attacks.
- Explainable AI techniques enhance the trustworthiness of the detection framework.
- Graph-aware learning is crucial for recognizing complex traffic patterns in IoT environments, leading to improved security.