Scholarlink

In this paper, we introduce Multi-Layer-GraphSAGE algorithm, a modified version of the GraphSAGE model tailored specifically for detecting attacks in heterogeneous datasets. With the proliferation of cyber threats across diverse network environments, there is a pressing need for robust and adaptable detection mechanisms. We conduct extensive experiments using multiple datasets. By prioritizing graph topologies over features, ML-GraphSAGE overcomes frequent obstacles in intrusion detection systems (IDS), such as difficulty recognizing minority classes in imbalanced datasets. The experimental results show that our proposed technique performs well across a variety of datasets. It achieves an average accuracy of 94.33% on the CIC-DDoS2019 dataset with 13 classes; 94.68% on the InSDN dataset with 8 classes; and 95.59% on the Edge-IIoT dataset with 15 classes.

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

Leveraging Graph Neural Network for Attack Detection Systems in Heterogeneous Network Data

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Samia Saidane,
  • Francesco Telch,
  • Kussai Shahin,
  • Fabrizio Granelli

摘要

In this paper, we introduce Multi-Layer-GraphSAGE algorithm, a modified version of the GraphSAGE model tailored specifically for detecting attacks in heterogeneous datasets. With the proliferation of cyber threats across diverse network environments, there is a pressing need for robust and adaptable detection mechanisms. We conduct extensive experiments using multiple datasets. By prioritizing graph topologies over features, ML-GraphSAGE overcomes frequent obstacles in intrusion detection systems (IDS), such as difficulty recognizing minority classes in imbalanced datasets. The experimental results show that our proposed technique performs well across a variety of datasets. It achieves an average accuracy of 94.33% on the CIC-DDoS2019 dataset with 13 classes; 94.68% on the InSDN dataset with 8 classes; and 95.59% on the Edge-IIoT dataset with 15 classes.