Answers are grounded in this paper and cite their sources.
Most existing encrypted traffic classification methods treat each traffic flow as an independent sample, overlooking the relationships shared by similar flows. This can amplify within-class variation and increase class overlap, especially when distinguishing characteristics are subtle and labeled data are limited. To address this problem, we propose PGCL-CG, a prototype-guided semi-supervised graph contrastive learning framework. We represent flow-level statistical features as a sparse, weighted k-nearest-neighbor graph and use GraphSAGE to aggregate information from neighboring flows and to capture their underlying structural relationships. We pretrain the encoder using self-supervised graph contrastive learning to obtain stable representations. We use a student–teacher architecture to produce more stable pseudo-labels and apply hierarchical density-based spatial clustering to filter out low-confidence assignments. We conduct extensive experiments on the two open datasets, demonstrating that PGCL-CG outperforms competitive baselines and remains robust even when only a small fraction of labels (e.g., 20% or 40%) is available. In particular, with only 5% of the training flows labeled, PGCL-CG achieves F1 scores of 0.820 and 0.771 on the ISCX-VPN and ISCX-nonVPN subsets, respectively, demonstrating its effectiveness under limited supervision.
No takes yet. Share an insight, caveat, or question.
Xin et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: