IDS-Anta: An open-source code with a defense mechanism to detect adversarial attacks for intrusion detection system
Kousik Barik, Sanjay Misra
Abstract
An intrusion detection system (IDS) is critical in protecting organizations from cyber threats. The susceptibility of Machine Learning and Deep Learning-based IDSs against adversarial attacks arises from malicious actors’ deliberate construction of adversarial samples. This study proposes a Python-based open-source code repository named IDS-Anta with a robust defence mechanism to identify adversarial attacks without compromising IDS performance. It uses Multi-Armed Bandits with Thomson Sampling, Ant Colony Optimization (ACO), and adversarial attack generation methods and is validated using three public benchmark datasets. This code repository can be readily applied and replicated on IDS datasets against adversarial attacks.