MTC-NET: A Multi-Channel Independent Anomaly Detection Method for Network Traffic.

Area of Science:

• Computer Science
• Cybersecurity
• Artificial Intelligence

Background:

• Deep learning, especially Transformer networks, shows promise for network traffic anomaly detection.
• Directly applying Transformers to noisy network traffic data degrades performance due to interference.
• Existing methods struggle with capturing long-distance dependencies and handling high-dimensional, noisy network traffic data.

Purpose of the Study:

• To propose MTC-Net, a novel multi-channel network traffic anomaly detection model.
• To reduce computational complexity and enhance the capture of long-distance dependencies in network traffic data.
• To improve the performance of anomaly detection in noisy network environments.

Main Methods:

• Decomposing network traffic sequences into multiple unidimensional time sequences.
• Employing a patch-based strategy to retain local semantic information within sub-sequences.
• Utilizing a hybrid backbone network combining Transformer and Convolutional Neural Network (CNN) architectures.
• Fusing channel information at the final classification header for comprehensive pattern modeling.

Main Results:

• MTC-Net demonstrates superior performance compared to state-of-the-art methods.
• The model achieves high accuracy, precision, recall, and F1 scores.
• Experiments conducted on KDD Cup 99, NSL-KDD, UNSW-NB15, and CIC-IDS2017 datasets validate the effectiveness of MTC-Net.

Conclusions:

• MTC-Net offers an effective solution for network traffic anomaly detection, particularly in noisy datasets.
• The multi-channel approach and hybrid backbone enhance the model's ability to detect complex network traffic patterns.
• The proposed method represents a significant advancement in the field of network security and anomaly detection.