Knowledge-to-Data: LLM-driven structured network traffic synthesis for testbed-free IDS evaluation
Knowledge-to-Data: LLM-Driven Synthesis of Structured Network Traffic for Testbed-Free IDS Evaluation
AI summary
The Knowledge-to-Data method constrains an LLM with protocol documents, attack semantics, and explicit statistical rules, synthesizing structured network traffic datasets without fine-tuning or real samples. In the AWID3 IEEE 802.11 benchmark case, datasets generated by 4 SOTA LLMs enabled gradient-boosted classifiers to achieve an F1 score of up to 0.956 when evaluated on real samples. The method provides a testbed-free, privacy-preserving, on-demand experimental alternative for IDS research.