Abstract: The evolution of cybersecurity threats demands efficient and accu-
rate attack detection systems, yet the scarcity of labeled data limits the use of
conventional supervised models. This paper proposes a Semi-Supervised Quan-
tum Generative Adversarial Network (sQGAN) for attack detection, combining
semi-supervised learning with quantum adversarial architectures to leverage la-
beled and unlabeled data for improved detection in data-scarce scenarios. Key
contributions include (1) a semi-supervised quantum architecture effective with
limited labeled data, (2) integration of quantum-based generator and discri-
minator networks to enhance attack detection, and (3) an experimental study
comparing sQGAN’s performance with quantum architectures. Results indicate
that sQGAN offers a high F1 score and robustness in detecting attacks under
challenging labeling conditions.