Network Traffic Classification Using Self-Supervised Learning and Confident Learning

2025-09-27

AI Summary

The rapid proliferation of encrypted traffic and dynamic port usage has rendered conventional network traffic classification (NTC) methods ineffective, while supervised learning approaches suffer from heavy reliance on large-scale labeled data and unsupervised methods yield insufficient accuracy. To address these challenges, this paper proposes a novel framework integrating self-supervised learning with confident learning. Specifically, it employs an autoencoder to extract traffic representations and introduces Tabular Contrastive Learning (TabCL)—a network-traffic-optimized contrastive learning scheme—to generate high-quality pseudo-labels. Confident learning (CL) is further incorporated to explicitly model and rectify label noise in the pseudo-labels. Extensive experiments on three real-world datasets demonstrate that the proposed method achieves substantial improvements in classification accuracy under extremely low labeling costs, outperforming existing state-of-the-art approaches. The framework exhibits high accuracy, strong generalization across diverse traffic scenarios, and practical deployability in real-world network environments.

Technology Category

• Machine Learning: Multi-class/Multi-label Learning & Extreme Classification
• Search and Optimization: Learning to Search
• Natural Language Processing: Safety and Robustness

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Web Mining and Content Analysis: Normalization, clustering, classification, and summarization of Web text
• Security and Privacy: Large-scale security measurements

Abstract

Network traffic classification (NTC) is vital for efficient network management, security, and performance optimization, particularly with 5G/6G technologies. Traditional methods, such as deep packet inspection (DPI) and port-based identification, struggle with the rise of encrypted traffic and dynamic port allocations. Supervised learning methods provide viable alternatives but rely on large labeled datasets, which are difficult to acquire given the diversity and volume of network traffic. Meanwhile, unsupervised learning methods, while less reliant on labeled data, often exhibit lower accuracy. To address these limitations, we propose a novel framework that first leverages Self-Supervised Learning (SSL) with techniques such as autoencoders or Tabular Contrastive Learning (TabCL) to generate pseudo-labels from extensive unlabeled datasets, addressing the challenge of limited labeled data. We then apply traffic-adopted Confident Learning (CL) to refine these pseudo-labels, enhancing classification precision by mitigating the impact of noise. Our proposed framework offers a generalizable solution that minimizes the need for extensive labeled data while delivering high accuracy. Extensive simulations and evaluations, conducted using three datasets (ISCX VPN-nonVPN, self-generated dataset, and UCDavis--QUIC), and demonstrate that our method achieves superior accuracy compared to state-of-the-art techniques in classifying network traffic.

Problem

• Addresses encrypted traffic classification challenges with limited labeled data
• Leverages self-supervised learning to generate pseudo-labels from unlabeled datasets
• Refines pseudo-labels using confident learning to enhance classification accuracy

Innovation

• Uses self-supervised learning for pseudo-label generation
• Applies confident learning to refine noisy labels
• Minimizes labeled data need while ensuring high accuracy