Few-shot traffic classification based on autoencoder and deep graph convolutional networks.

Area of Science:

• Computer Science
• Network Engineering
• Machine Learning

Background:

• Network traffic classification is vital for network management, optimizing efficiency, QoS, security, and policy enforcement.
• Graph Convolutional Networks (GCNs) are increasingly used for traffic classification, considering data features and relationships.
• Existing GCN methods often use shallow architectures (two-layer) to avoid over-smoothing, limiting performance on small datasets.

Purpose of the Study:

• To propose a novel method, Autoencoder and Deep Graph Convolutional Networks (ADGCN), for traffic classification in few-shot learning scenarios.
• To address the limitations of zero-padding in traffic data preprocessing for GCNs.
• To enhance the classification performance of GCNs when dealing with limited traffic samples.

Main Methods:

• Utilized an autoencoder (AE) to reconstruct traffic data, learning abstract features to mitigate zero-padding effects.
• Employed GCNII, a deep GCN model, for classifying the reconstructed traffic, designed to handle insufficient data samples.
• Developed an end-to-end ADGCN framework applicable to various traffic classification scenarios.

Main Results:

• The proposed ADGCN method demonstrated significant improvements in classification accuracy, ranging from 3.5% to 24% compared to state-of-the-art approaches.
• The AE component effectively addressed the adverse effects of zero-padding on traffic classification with small samples.
• The deep GCN architecture (GCNII) proved effective in capturing complex relationships in limited traffic data.

Conclusions:

• ADGCN offers a robust and effective solution for network traffic classification, particularly in scenarios with limited data.
• The integration of autoencoders and deep GCNs overcomes key challenges in current GCN-based traffic classification methods.
• The method shows promising results, advancing the field of network traffic analysis and management.