Evaluating Synthetic Malicious Network Traffic Generated by GAN and VAE Models: A Data Quality Perspective

The limited availability and imbalance of labeled malicious network traffic data remain major obstacles in developing effective AI-driven cybersecurity solutions. To mitigate these challenges, this study investigates the use of deep generative models, specifically Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs), for producing realistic synthetic attack data. A comprehensive data quality assessment (DQA) framework is proposed to thoroughly evaluate the fidelity, diversity, and practical utility of the generated data samples. The findings support the adoption of data synthesis as a viable strategy to address data scarcity, improving robustness and reliability in modern cybersecurity applications and sectors.