Scholarlink

<p>This paper proposed an innovative defensive approach against Advanced Persistent Threats (APTs), combining feature selection via Generative Adversarial Networks (GANs) with deep learning-based graph sampling and aggregation (GraphSAGE). APTs refer to remarkably sophisticated cyberattacks, frequently related to cyberwarfare and industrial espionage. These threats use complex strategies to penetrate intelligent systems, making them particularly difficult to detect. This framework addressed this gap by: (a) dynamically selecting critical features using GANs, reducing noise, and highlighting attack patterns; (b) modeling topological dependencies via GraphSAGE to detect stealthy APT behaviors; (c) validating the approach on European datasets (ELECTRON, Cyber4OT) containing more than 2 million packets; (d) conducting rigorous hyperparameter optimization. This methodology demonstrated the effectiveness of the proposed approach, achieving a remarkable detection accuracy of 94.45%, outperforming the state-of-the-art methods by 0.45%, while preserving resource efficiency (17% memory, 65% CPU utilization). Thus, this work establishes a scalable and adaptive defense for constantly evolving APTs, strengthening the resilience of smart grids.</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

GAN-Driven Feature Selection and GraphSAGE for Advanced Persistent Threat Defense in Smart Grids

  • 期刊论文

  • 发表日期：

    2025年9月18日

查看全文
  • Lahcen Hassine,
  • Hasna Chaibi,
  • Mohamed Rahouti,
  • Rachid Saadane,
  • Abdellah Chehri,
  • Adil Mehdary

摘要

This paper proposed an innovative defensive approach against Advanced Persistent Threats (APTs), combining feature selection via Generative Adversarial Networks (GANs) with deep learning-based graph sampling and aggregation (GraphSAGE). APTs refer to remarkably sophisticated cyberattacks, frequently related to cyberwarfare and industrial espionage. These threats use complex strategies to penetrate intelligent systems, making them particularly difficult to detect. This framework addressed this gap by: (a) dynamically selecting critical features using GANs, reducing noise, and highlighting attack patterns; (b) modeling topological dependencies via GraphSAGE to detect stealthy APT behaviors; (c) validating the approach on European datasets (ELECTRON, Cyber4OT) containing more than 2 million packets; (d) conducting rigorous hyperparameter optimization. This methodology demonstrated the effectiveness of the proposed approach, achieving a remarkable detection accuracy of 94.45%, outperforming the state-of-the-art methods by 0.45%, while preserving resource efficiency (17% memory, 65% CPU utilization). Thus, this work establishes a scalable and adaptive defense for constantly evolving APTs, strengthening the resilience of smart grids.