Answers are grounded in this paper and cite their sources.
With the advancement of network communication technology, open-set network traffic classification has become increasingly important. Existing traffic classification methods are not sufficiently robust and face two major challenges: first, closed-set network traffic classification methods, which focus solely on known classes, cannot handle open-set scenarios; second, existing open-set network traffic classification methods rely on the ideal assumption that the class sample distribution in the training set is uniform to achieve strong open-set performance. However, this assumption is overly idealistic and significantly limits the applicability of such methods in real-world scenarios. To address this issue, we propose RoeCi, an open-set network traffic classification method designed for scenarios with imbalanced sample distribution of training classes in real-world environments. RoeCi integrates uncertainty estimation with multi-expert models to mitigate this imbalance. We construct a sample complexity metric for each flow using the second-order classification probability, which guides the adaptive deployment of expert models with varying numbers and capabilities, as well as the dynamic weighting of their parameters. Furthermore, this complexity metric is also used to address the open-set classification problem involving both known and unknown flows. Finally, we evaluate the effectiveness of the proposed method on three public datasets of known classes and six open-set scenarios. RoeCi achievesF1that are 12.08%–61.28% higher than state-of-the-art methods on known classes and 23.23%–67.80% higher in open-set scenarios.
No takes yet. Share an insight, caveat, or question.
Wang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: