Loading...

Analyzing Robustness of Automated Intrusion Detection Systems in Local Networks

Dadkhah Tehrani, Pouria | 2025

0 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 58219 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Aref, Mohammad Reza; Ahmadi, Siavash
  7. Abstract:
  8. With the growing prevalence of cyber-attacks and the increasing complexity of communication networks, network security has become one of the most critical challenges in the field of information technology. Among the most effective tools for securing networks are Intrusion Detection Systems (IDS), which are capable of identifying abnormal behaviors and network-based attacks. However, many IDSs are vulnerable to adversarial attacks—carefully crafted manipulations intended to mislead their detection mechanisms. This thesis proposes an enhanced adversarial attack framework that leverages Generative Adversarial Networks (GANs), attention mechanisms, and active learning to generate deceptive network traffic capable of evading IDS detection. The proposed approach distinguishes between functional and non-functional features of network traffic, allowing the GAN to focus on producing adversarial packets that maintain operational validity while deceiving IDS models. The attention mechanism enables the model to emphasize critical features during training, and the active learning component significantly reduces the number of required queries, thereby lowering computational costs. Experimental results on a real-world intrusion detection dataset demonstrate that the proposed model can reduce the accuracy of various IDS algorithms by up to 50% on average. This highlights the model’s effectiveness in generating realistic and impactful adversarial attacks. Moreover, the integration of attention and active learning mechanisms improves efficiency and reduces the dependency on computational resources. Beyond its ability to degrade IDS performance, the proposed framework can serve as a diagnostic tool to uncover weaknesses in intrusion detection systems and support the development of more robust network defense strategies. This research introduces a novel adversarial technique and opens new directions for future work in adversarial machine learning and the design of resilient security systems
  9. Keywords:
  10. Adversarial Attacks ; Intrusion Detection System ; Generative Adversarial Networks ; Attention Mechanism ; Active Learning ; Network Security

 Digital Object List

 Bookmark

No TOC