Scholarlink

<p>The increasing complexity and diversity of cyber-attacks in interconnected systems, including the Internet of Things (IoT), industrial control networks, and critical infrastructure, have exposed the limitations of traditional intrusion detection methods based on predefined rules or static architectures. To address these challenges, we propose a novel hybrid intrusion detection framework that synergistically integrates Graph Neural Networks (GNNs) and Transformer architectures through a dual-domain feature fusion strategy. Specifically, network traffic data are converted into grayscale images, from which texture features are extracted via the Gray Level Co-occurrence Matrix (GLCM) method and fused with traffic-based features to enhance spatial–temporal representation. Borderline-Synthetic Minority Over-sampling Technique (SMOTE) is employed to mitigate class imbalance, improving robustness against minority-class misclassification. Furthermore, a Kolmogorov–Arnold Network (KAN) replaces the conventional multilayer perceptron (MLP) head to capture richer non-linear decision boundaries. Extensive experiments on UNSW_NB15, IEC61850, and DAPT2020 datasets show that our approach consistently outperforms state-of-the-art baselines in accuracy, achieving up to 99.24% while maintaining competitive efficiency. The proposed framework offers a scalable and generalizable solution for intrusion detection across diverse network environments, contributing to the advancement of resilient cybersecurity systems.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A novel hybrid graph neural network and transformer model for intrusion detection

  • 期刊论文

  • 发表日期：

    2026年3月6日

查看全文
  • Rong Xie,
  • Die Liu

摘要

The increasing complexity and diversity of cyber-attacks in interconnected systems, including the Internet of Things (IoT), industrial control networks, and critical infrastructure, have exposed the limitations of traditional intrusion detection methods based on predefined rules or static architectures. To address these challenges, we propose a novel hybrid intrusion detection framework that synergistically integrates Graph Neural Networks (GNNs) and Transformer architectures through a dual-domain feature fusion strategy. Specifically, network traffic data are converted into grayscale images, from which texture features are extracted via the Gray Level Co-occurrence Matrix (GLCM) method and fused with traffic-based features to enhance spatial–temporal representation. Borderline-Synthetic Minority Over-sampling Technique (SMOTE) is employed to mitigate class imbalance, improving robustness against minority-class misclassification. Furthermore, a Kolmogorov–Arnold Network (KAN) replaces the conventional multilayer perceptron (MLP) head to capture richer non-linear decision boundaries. Extensive experiments on UNSW_NB15, IEC61850, and DAPT2020 datasets show that our approach consistently outperforms state-of-the-art baselines in accuracy, achieving up to 99.24% while maintaining competitive efficiency. The proposed framework offers a scalable and generalizable solution for intrusion detection across diverse network environments, contributing to the advancement of resilient cybersecurity systems.