A Review of Generative Models in Generating Synthetic Attack Data for Cybersecurity

The ability of deep learning to process vast data and uncover concealed malicious patterns has spurred adoption in cybersecurity, but privacy and security constraints limit access to sufficiently large, realistic datasets. As an alternative, researchers increasingly turn to synthetic data generation, notably with generative adversarial networks (GANs). While GANs are widely used across domains, their efficacy in producing realistic cyberattack data and the performance of downstream models trained on such data remain open questions. This paper reviews generative learning fundamentals, examines GAN architectures and data-generation capabilities, surveys GAN-based methods for synthesizing cyberattack data, and empirically analyzes the fidelity and utility of GAN-generated attack data using NSL-KDD, with an emphasis on DoS attacks. The study aims to clarify the potential of synthetic data to bolster deep learning for robust cybersecurity.