An encrypted traffic classification method based on autoencoders and convolutional neural networks.

Area of Science:

• Computer Science
• Network Security
• Machine Learning

Background:

• Existing encrypted traffic classification methods face challenges with large datasets, high computational costs, and poor generalization.
• Zero-padding in traffic classification can negatively impact accuracy when uniform flow lengths are used.

Purpose of the Study:

• To propose an effective encrypted traffic classification method addressing limitations of existing approaches.
• To improve the accuracy and robustness of encrypted traffic classification.

Main Methods:

• Utilized an autoencoder for dataset reconstruction, enabling effective classification with smaller datasets.
• Employed autoencoders to learn abstract feature representations from traffic flows, mitigating zero-padding effects.
• Applied convolutional neural networks (CNNs) for traffic classification, leveraging their strong generalization capabilities.

Main Results:

• Achieved a classification accuracy improvement of 2.86% to 18.13% compared to existing advanced methods.
• Demonstrated enhanced robustness in traffic classification compared to other state-of-the-art methods.
• Successfully enabled effective classification with smaller-scale datasets.

Conclusions:

• The proposed autoencoder and CNN-based method offers a superior solution for encrypted traffic classification.
• The method effectively overcomes the limitations of traditional approaches, providing higher accuracy and robustness.
• The approach is particularly beneficial when dealing with limited training data and aims to mitigate issues associated with zero-padding.