A lightweight anomaly detection model for network traffic using multi scale spatio temporal residual learning.

Area of Science:

• Cybersecurity
• Network Security
• Machine Learning

Background:

• Network attacks are evolving, necessitating advanced abnormal traffic detection.
• Existing methods struggle with large-scale data, class imbalance, and computational efficiency.

Purpose of the Study:

• To develop a lightweight knowledge transfer anomaly detection network for improved network security.
• To address challenges in large-scale traffic analysis, class imbalance, and computational efficiency.

Main Methods:

• Utilized multi-scale residual networks for spatio-temporal feature extraction.
• Implemented a lightweight knowledge transfer anomaly detection network with knowledge distillation.
• Migrated knowledge from a teacher model to a lightweight student model.

Main Results:

• Achieved 0.93 accuracy and 0.27 loss at 100 iterations.
• Reached 0.97 specificity with a sample size of 5000.
• Demonstrated low training (22.8s) and inference (0.06s) times.
• Showcased 0.07% traffic loss under high attack intensity.

Conclusions:

• The proposed method effectively handles complex network traffic data.
• Enhanced precision and deployment effectiveness for anomalous detection.
• Shows significant promise for real-world network security applications.