Revolutionizing Encrypted Traffic Classification with MH-Net: A Multi-View Heterogeneous Graph Model
Authors: Haozhen Zhang, Haodong Yue, Xi Xiao, Le Yu, Qing Li, Zhen Ling, Ye Zhang
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments conducted on the ISCX and CIC-Io T datasets for both the packet-level and flow-level traffic classification tasks demonstrate that MH-Net achieves the best overall performance compared to dozens of SOTA methods. |
| Researcher Affiliation | Academia | 1Tsinghua Shenzhen International Graduate School, Tsinghua University, Shenzhen, China 2Peng Cheng Laboratory, Shenzhen, China 3Key Laboratory of Cyberspace Security, Ministry of Education of China, Zhengzhou, China 4Nanjing University of Posts and Telecommunications, Nanjing, China 5Southeast University, Nanjing, China 6National University of Singapore, Singapore EMAIL, EMAIL, EMAIL, EMAIL, EMAIL, EMAIL, EMAIL |
| Pseudocode | No | The paper describes the methodology in detail using prose and mathematical formulations, but it does not contain any explicit pseudocode blocks or algorithms. |
| Open Source Code | Yes | Code https://github.com/Viktor Axelsen/MH-Net |
| Open Datasets | Yes | To thoroughly evaluate MH-Net on the packetlevel and flow-level traffic classification tasks, we adopt the CIC-Io T (Dadkhah et al. 2022), ISCX VPN-non VPN (Gil et al. 2016) and ISCX Tor-non Tor (Lashkari et al. 2017) datasets. |
| Dataset Splits | Yes | we first adopt stratified sampling to partition the flow-level training and testing dataset into 9:1 according to the number of traffic flows for all datasets. All packets in the flow-level training and testing datasets are directly used as the packet-level training and testing datasets, respectively. |
| Hardware Specification | Yes | The experimental results are reported as the mean over five runs on an NVIDIA RTX 3080. |
| Software Dependencies | No | The paper states: "We implement MH-Net and conduct all experiments with Py Torch and Deep Graph Library." However, it does not specify version numbers for PyTorch or Deep Graph Library. |
| Experiment Setup | Yes | The max flow length (i.e., the max packet number within a flow) is set to 15. The number of layers in HGNN is set to 4, and we initialize the RNN to LSTM in MH-Net by default. As for the random walk, we set the scale of subgraphs following (Qiu et al. 2020), and the restart probability is 0.8. We set the packet dropping ratio PPD to 0.6, and the temperature coefficient τ is 0.07. The objective coefficients α and β are set to 1.0 and 0.5, respectively. |