Synthetic Encrypted Traffic Generation with VAE-GAN: Capturing Complex Patterns for Realistic Network Modeling

The growing volume of encrypted traffic in modern communication networks has underscored the need to preserve privacy while enhancing network performance. Traffic generation techniques are crucial for achieving these objectives. Traditional methods of network traffic generation have been limited in terms of generating realistic and complex traffic patterns. Recent advances in machine learning have now enabled the development of new solutions to these challenges. We introduce one such solution in this work that utilizes the power of Variational Autoencoders (VAEs) and Generative Adversarial Networks (GANs) to generate synthetic traffic data. Our proposed VAE-GAN model is expected to reproduce the intricate patterns of encrypted traffic, i.e., rare bursts and chunks, which are crucial for network performance testing. To evaluate the fidelity of generated traffic, we apply statistical and distance-based methods. Statistical methods analyze the distributions of the original and generated data, whereas distance-based methods, such as Euclidean and Westessin distances, compute the similarity between the two datasets. Our experimental results indicate that the VAE-GAN model effectively synthesizes traffic very similar to the original data, with a maximum distance value of 0.3 seen between generated traffic and original traffic.