Balancing Complexity and Performance in Convolutional Neural Network Models for QUIC Traffic Classification.

Area of Science:

• Computer Science
• Telecommunications Engineering
• Network Security

Background:

• Sixth-generation (6G) wireless networks demand advanced Network Traffic Classification (NTC) for performance and security.
• Resource-constrained Internet of Things (IoT) devices require efficient traffic analysis and threat detection.
• Encrypted communications, like Quick UDP Internet Connections (QUIC), challenge traditional NTC methods.

Purpose of the Study:

• To investigate the effectiveness of Convolutional Neural Networks (CNNs) for NTC with encrypted traffic.
• To evaluate the trade-off between classification accuracy and CNN model complexity.
• To demonstrate practical NTC solutions for 6G and IoT environments.

Main Methods:

• Utilized statistical analysis of network flow characteristics.
• Developed and assessed various CNN architectures for NTC.
• Focused on encrypted traffic, specifically from the QUIC protocol.
• Compared performance against existing baseline architectures.

Main Results:

• Achieved nearly 92% accuracy in classifying encrypted network traffic using CNNs.
• Demonstrated that low-complexity CNN models offer a favorable balance between accuracy and computational efficiency.
• Showcased the viability of ML-based NTC for modern, encrypted network environments.

Conclusions:

• CNNs are effective for NTC, even with encrypted traffic like QUIC.
• Simple, low-complexity CNN architectures provide a practical solution for efficient network analysis.
• This research supports the development of secure and high-performance 6G and IoT networks.