← Back to Publications List

Evaluating and Hardening DDoS Defenses under Adaptive Adversaries Using Attack Success Rate

Students & Supervisors

Student Authors
Faishal Uddin Himel
Bachelor of Science in Computer Science & Engineering, FST
Nasmin Suraya
Bachelor of Science in Computer Science & Engineering, FST
Supervisors
Kamruddin Nur
Professor, Faculty, FST

Abstract

Distributed Denial-of-Service (DDoS) attacks have kept progressing in terms of size and complexity, posing challenges to the effectiveness of existing machine learning-based DDos detection systems. Traditional methods assume the usage of synthetic balanced data, evaluate using IID, and consider static attacks which makes them less effective in practice. In this paper, an adversarial-resilient DDoS defense system is presented which maintains traffic imbalance through imbalanced training and cost-sensitive learning. A strict cross-attack evaluation strategy is employed for evaluation of generalization ability to new attack families while the Attack Success Rate (ASR) metric is used to evaluate the resilience against adaptive attacks. To mimic the adaptive attackers, a Generative Adversarial Network (GAN) creates an adaptive DDoS attack and then the model is trained adversarially to increase its resilience. Both SHAP and LIME are included to ensure that decisions of the model are based on legitimate features of traffic.

Keywords

DDoS defence adversarial machine learning attack success rate GAN-based attacks and cost-sensitive learning.

Publication Details

  • DOI: doi: 10.1109/QPAIN69676.2026.11546153
  • Type of Publication:
  • Conference Name: 2026 IEEE 2nd International Conference on Quantum Photonics Artificial Intelligence and Networking (QPAIN)
  • Date of Conference: 16/04/2026 - 16/04/2026
  • Venue: Chittagong, Bangladesh
  • Organizer: IEEE Photonics Society Bangladesh Chapter
View on Publisher