A Novel Lightweight Anonymous Proxy Traffic Detection Method Based on Spatio-Temporal Features.

Area of Science:

• Computer Science
• Cybersecurity
• Network Security

Background:

• Anonymous proxies facilitate illegal network activities like data theft and cyber attacks.
• Deep learning methods are increasingly used for network traffic detection due to automated feature extraction.
• Existing image-based deep learning approaches for traffic detection suffer from large storage and computational overhead.

Purpose of the Study:

• To propose a novel method for anonymous proxy traffic detection that reduces storage and computational resource overhead.
• To address the limitations of large-sized image representations in current deep learning-based traffic detection methods.

Main Methods:

• Converting sequences of packet size and inter-arrival time of the first N packets into images.
• Utilizing a one-dimensional convolutional neural network (1D-CNN) for categorizing the converted images.
• Validating the method using both proprietary and public datasets.

Main Results:

• The proposed method generates network traffic images that are at least 90% smaller than those from existing image-based methods.
• Achieved high detection accuracy with F1 scores up to 98.51% for Shadowsocks traffic and 99.8% for VPN traffic.
• Demonstrated the effectiveness of the method in reducing resource consumption while maintaining detection performance.

Conclusions:

• The novel image conversion technique significantly reduces the overhead associated with deep learning-based network traffic analysis.
• The method offers an efficient and effective solution for anonymous proxy traffic detection, crucial for enhancing network security.
• This approach provides a viable alternative for real-time network security monitoring and threat mitigation.