Deep Ensemble Learning for Application Traffic Classification Using Differential Model Selection Technique.

Area of Science:

• Computer Science
• Network Engineering
• Artificial Intelligence

Background:

• Modern internet traffic is increasingly complex, necessitating advanced application traffic classification methods.
• Deep learning models show promise but face challenges in balancing accuracy, inference time, and generalization.
• Traditional heuristic methods struggle with the diversity of current network traffic.

Purpose of the Study:

• To develop an end-to-end learning method for improved application traffic classification.
• To enhance the performance-inference time trade-off in network traffic classifiers.
• To address the limitations of existing deep learning and heuristic approaches.

Main Methods:

• An end-to-end learning framework was developed.
• A model-selection-based ensemble mechanism was incorporated.
• The method was evaluated on public and private network traffic datasets.

Main Results:

• The proposed method demonstrated improved classification accuracy across all tested datasets.
• The approach maintained reasonable inference times compared to existing methods.
• Performance gains were observed against nine other classification techniques.

Conclusions:

• The proposed method offers a superior performance-inference time trade-off for application traffic classification.
• This approach effectively handles diverse and complex internet traffic patterns.
• The model-selection-based ensemble mechanism is key to achieving high accuracy and efficiency.