Scholarlink

<p>With adversarial attacks against IDS becoming increasingly sophisticated, new methods are needed to stress-test their vulnerabilities. While GANs have been used for adversarial attacks in other domains, their application to deceive low-rate DDoS-IDS remains largely unexplored. This paper proposes a framework that leverages GANs to generate adversarial examples specifically optimized to evade DDoS-IDS models. We train IDS models on two benchmark datasets (the first dataset: Low-Rate DDoS Dataset 2022, and the second Dataset: “Public dataset for evaluating Port Scan and Slowloris attacks”), then design a GAN-based attack achieving a 99.9% evasion rate. Our experiments reveal a critical architectural flaw in modern DDoS-IDS: even minimal perturbations to low-rate traffic bypass detection. This work establishes a new threat model for DDoS-IDS and underscores the need for adversarial-resistant designs.</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

Resilient intrusion detection system for adversarial attacks on Low-Rate DDoS

  • 期刊论文

  • 发表日期：

    2025年7月12日

查看全文
  • Qasem Abu Al-Haija,
  • Ayat Droos

摘要

With adversarial attacks against IDS becoming increasingly sophisticated, new methods are needed to stress-test their vulnerabilities. While GANs have been used for adversarial attacks in other domains, their application to deceive low-rate DDoS-IDS remains largely unexplored. This paper proposes a framework that leverages GANs to generate adversarial examples specifically optimized to evade DDoS-IDS models. We train IDS models on two benchmark datasets (the first dataset: Low-Rate DDoS Dataset 2022, and the second Dataset: “Public dataset for evaluating Port Scan and Slowloris attacks”), then design a GAN-based attack achieving a 99.9% evasion rate. Our experiments reveal a critical architectural flaw in modern DDoS-IDS: even minimal perturbations to low-rate traffic bypass detection. This work establishes a new threat model for DDoS-IDS and underscores the need for adversarial-resistant designs.