Multi-Task Scenario Encrypted Traffic Classification and Parameter Analysis.

Area of Science:

• Cybersecurity
• Network Analysis
• Machine Learning

Background:

• Encrypted traffic presents significant challenges for network management and security.
• Traditional machine learning methods struggle with the increasing volume and complexity of encrypted data.
• Deep learning offers improved accuracy but often requires substantial computational resources.

Purpose of the Study:

• To develop a more efficient deep learning method for encrypted traffic classification.
• To address limitations in computational memory consumption and interpretability of existing models.
• To enhance network security and analysis capabilities through improved encrypted traffic classification.

Main Methods:

• Introduced a Parameter-Efficient Fine-Tuning (PEFT) method for encrypted traffic classification models.
• Conducted experiments on diverse public datasets for Tor traffic service and malicious traffic classification.
• Performed fair comparisons against state-of-the-art deep learning architectures.

Main Results:

• The proposed PEFT method significantly reduced the scale of fine-tuning parameters and computational resource usage.
• Achieved performance comparable to existing state-of-the-art deep learning models.
• Interpreted the model's learning mechanism, revealing a hierarchical structure and distinct feature representation.

Conclusions:

• Parameter-Efficient Fine-Tuning offers an effective solution for efficient and accurate encrypted traffic classification.
• The method enhances network security and analysis by optimizing deep learning model performance.
• The study validates the effectiveness and interpretability of the proposed approach for encrypted traffic analysis.