Network Traffic Generation by Large Language Models (LLMs)

This repository presents a novel approach for generating realistic network traffic using Large Language Models (LLMs), specifically OpenAI’s GPT-4.1 and GPT-5.

Our method, called the Large Language Model Network Traffic Generator (LLM-NTG), aims to bridge the gap between realistic traffic generation and the expressive capabilities of LLMs.
We employ a few-shot learning framework combined with a human-in-the-loop feedback mechanism, where generated traffic is continuously evaluated and refined.

---

📂 Repository Structure

1. Datasets/

Contains datasets required for traffic generation:

• One-way and two-way traffic datasets.
• Sample inputs in .json, .pcapng, and .csv formats for traffic generation.

---

2. GPT-4.1/

Experiments performed with GPT-4.1.
This folder has two subfolders: Experiment_1/ and Experiment_2/.

Each experiment includes:

• Exp1_Sample_Packets_Extraction.ipynb
  Extracts traffic data from the dataset and prepares sample packets for generation.
• Exp1_Traffic_Generation_gpt4.1.ipynb
  Generates synthetic traffic and saves the output as .json files.
• Exp1_Statistics_of_Generated_Traffics_gpt4.1.ipynb
  Computes statistics of the generated traffic.

(Experiment_2/ follows the same structure.)

---

3. GPT-5/

Experiments performed with GPT-5.
This folder also has Experiment_1/ and Experiment_2/.

Since the same input samples are used as in GPT-4.1, there is no extraction notebook here.
Each experiment includes:

• Exp1_Traffic_Generation_gpt5.ipynb
  Generates synthetic traffic with GPT-5.
• Exp1_Statistics_of_Generated_Traffics_gpt5.ipynb
  Computes statistics of the generated traffic.

(Experiment_2/ follows the same structure.)

---

4. Generated_Traffic/

Contains the generated traffic outputs in .json format and .pcap files.

• Each experiment’s results are stored here.
• The JSON_files/ subfolder contains details such as: .json format of generated traffic
• The PCAP_files/ subfolder contains details such as: There are .pcap files of the generated traffi,c and Wireshark can be used to view them.
• The Results/ subfolder contains details such as:
  • Token usage
  • Computation time

---

5. pcap_converter.py

A transformation script that converts generated .json traffic files into .pcap format, enabling further analysis with standard network traffic tools (e.g., Wireshark).

---

6. PCAP_GPT4.1_Experiment_1.png and PCAP_GPT5_Experiment_1.png

Wireshark representation of traffic generated by GPT 4.1 and 5 for Experiment~1.

---

🚀 Usage

1. Explore the datasets under Datasets/.
2. Run the notebooks in GPT-4.1/ or GPT-5/ for traffic generation.
3. Generated traffic will be saved under Generated_Traffic/.
4. Optionally, convert .json traffic files into .pcap format using:

   python pcap_converter.py input.json output.pcap