FedETC: Encrypted traffic classification based on federated learning.

Area of Science:

• Computer Science
• Network Security
• Machine Learning

Background:

• Traditional traffic classification methods struggle with diverse tasks and require extensive feature engineering.
• Data privacy regulations restrict the collection and sharing of network traffic data, hindering centralized machine learning approaches.
• Existing solutions often fail to balance classification accuracy with user privacy preservation.

Purpose of the Study:

• To propose FedETC, a novel federated learning framework for network traffic classification.
• To enable collaborative learning of global traffic classifiers while preserving local data privacy.
• To address the limitations of manual feature engineering in traffic classification tasks.

Main Methods:

• FedETC framework utilizing federated learning principles.
• Integration of a one-dimensional convolutional neural network (1D-CNN) as the base model, eliminating the need for manual feature design.
• Evaluation on a publicly available real-world dataset for application identification and traffic characterization.

Main Results:

• FedETC achieves high accuracy in both application identification and traffic characterization tasks.
• The framework demonstrates performance comparable to centralized learning schemes.
• Local traffic data remains encrypted and invisible to other participants, ensuring privacy.

Conclusions:

• FedETC offers an effective solution for privacy-preserving network traffic classification.
• The framework successfully overcomes the challenges of feature engineering and data privacy in traffic analysis.
• Federated learning with 1D-CNN provides a viable alternative to traditional centralized methods.