A Comprehensive Survey on Network Traffic Synthesis: From Statistical Models to Deep Learning

The description provided in the query refers to the work by Almasre et al., not the survey itself. According to the survey A Comprehensive Survey on Network Traffic Synthesis: From Statistical Models to Deep Learning by Sivaroopan et al., Almasre et al. utilized Conditional Generative Adversarial Networks (CGANs) to generate IoT network traffic (including various cyberattacks) to improve the analysis of security vulnerabilities.

The survey itself, authored by Nirhoshan Sivaroopan, Thilini Dahanayaka, and Kanchana Thilakarathna (published June 2025 on arXiv), provides a comprehensive review of network traffic synthesis. It categorizes methods into statistical models (e.g., Markov Models, Poisson processes) and deep learning techniques (e.g., VAEs, GANs, Diffusion Models, Transformers), while also discussing data types, evaluation metrics, and open challenges.

This material provides a broad, technically oriented survey of network traffic synthesis, tracing the field’s evolution from classical statistical and probabilistic generative models to modern deep-learning approaches. It frames synthetic traffic generation as a core capability for cybersecurity research, where realistic network data is needed for intrusion detection, anomaly analysis, vulnerability assessment, and machine-learning training, but raw traffic is often sensitive, scarce, or dominated by benign examples. The survey organizes the literature around the goals and challenges of synthesis: preserving statistical structure, maintaining protocol-level realism, retaining attack signatures and labels, ensuring scalability, and avoiding privacy leakage or semantic distortion.

A key contribution is the move beyond purely distributional modeling toward ontology-guided synthesis. By leveraging a unified cybersecurity ontology, the work constructs a network-traffic knowledge base that connects traffic features, protocol behavior, attack classes, and vulnerability context. This semantic layer enables generative models—particularly conditional GANs—to produce targeted, class-conditioned traffic samples rather than only generic traffic-like sequences. In the IoT setting, this is especially valuable because attack traffic is often rare, difficult to collect ethically, and tightly coupled to specific device behaviors and exploit conditions. Conditioning synthesis on ontological structure helps generate more meaningful attack scenarios for vulnerability analysis and downstream detection tasks.

The material matters because it situates network traffic synthesis as both a methodological and an infrastructure-level problem in cybersecurity. For practitioners, it highlights how synthetic traffic can support data augmentation, privacy-preserving benchmarking, red-team simulation, and evaluation of detection systems without exposing sensitive production networks. Its emphasis on combining ontologies with deep generative models suggests a path toward more controllable, explainable, and security-relevant synthetic data, while also underscoring the need for rigorous evaluation metrics that capture realism, utility, label fidelity, and privacy risk rather than relying solely on surface-level distributional similarity.