CBD: A Deep-Learning-Based Scheme for Encrypted Traffic Classification with a General Pre-Training Method.

Area of Science:

• Computer Science
• Network Security
• Machine Learning

Background:

• Encrypted traffic is increasing, making traditional classification models difficult to apply in real environments due to reliance on labeled data.
• Existing encrypted traffic classification methods often lack generalization capabilities for real-world network conditions.
• Effective analysis of network traffic is crucial for security and performance monitoring.

Purpose of the Study:

• To propose a transferable model, CBD, for generalized encrypted traffic classification in real environments.
• To address the limitations of current models that require labeled data and struggle with real-world applicability.
• To enhance the ability to analyze and classify encrypted network traffic.

Main Methods:

• Developed a transferable model named CBD, integrating a one-dimension Convolutional Neural Network (CNN) and a Transformer encoder.
• Employed a pre-training strategy using unlabeled data to capture fundamental characteristics of encrypted traffic.
• Evaluated the model's performance on packet-level and flow-level classification using a public dataset.

Main Results:

• The proposed CBD model demonstrated superior performance compared to baseline methods in encrypted traffic classification.
• Pre-training on unlabeled data significantly enhanced the model's classification capabilities and generalization.
• The model effectively classifies encrypted traffic at both packet and flow levels.

Conclusions:

• The CBD model offers a robust solution for encrypted traffic classification in real-world network scenarios.
• Transfer learning and pre-training are effective strategies for improving the generalization of encrypted traffic classification models.
• This research contributes to more accurate and adaptable network traffic analysis in the face of increasing encryption.