A Systematic Survey of Generative Adversarial Networks (GAN) and Transformer Models for Zero-Day Intrusion Detection
Abstract
Intrusion Detection Systems (IDS) play a vital role in modern networks by monitoring network activities, identifying malicious behavior and mitigating potential cyber threats. Traditional Intrusion Detection Systems are plagued with novel hacking techniques, zero-day attacks and infiltrations. These threats strategically avoid signature-based and machine learning-based methods due to the scarcity of data, the class imbalance, and the limited abilities of generalization. To overcome these limitations, recent studies have explored the use of deep learning architectures such as Generative Adversarial Networks (GANs) and Transformer-based models. This paper presents a structured survey of GAN- and Transformer-based IDS models, highlighting their individual and combined potential for improving detection of rare and zero-day attacks. In such frameworks, GANs generate realistic synthetic attack samples to address class imbalance, while Transformer models capture long-range dependencies in network traffic for robust anomaly detection. The reviewed studies evaluated the approaches on benchmark datasets, including CICIDS2017, NSL-KDD, and UNSW-NB15, reporting improvements in detection accuracy and reductions in false-positive rates compared with traditional machine learning methods. Furthermore, this survey identifies key research gaps, challenges and future directions. The findings indicate that Generative Adversarial Networks (GANs) can support data augmentation, while Transformers can enhance anomaly detection, emphasizing the strength for improving intrusion detection capabilities.