Research on Anomaly Network Detection Based on Self-Attention Mechanism.

Area of Science:

• Computer Science
• Cybersecurity
• Machine Learning

Background:

• Network traffic anomaly detection is crucial for identifying and preventing security threats.
• Existing methods often face challenges with efficiency and accuracy in complex network environments.

Purpose of the Study:

• To develop a novel deep-learning-based model for enhanced network traffic anomaly detection.
• To improve the efficiency and accuracy of anomaly detection through advanced feature engineering and a specialized neural network architecture.

Main Methods:

• Reconstruction of a network traffic anomaly detection dataset (DNTAD) using enhanced feature extraction from UNSW-NB15.
• Development of a detection model integrating Long Short-Term Memory (LSTM) and a recurrent neural network self-attention mechanism to capture temporal dependencies and feature relationships.

Main Results:

• The proposed feature engineering method improved operational efficiency for classic machine learning algorithms like XGBoost without compromising training performance.
• The LSTM-based model with self-attention mechanism demonstrated superior performance compared to other models on the reconstructed dataset, validating the effectiveness of its components through ablation studies.

Conclusions:

• The study successfully developed and validated a new deep learning model for network traffic anomaly detection.
• The enhanced feature engineering and LSTM-self-attention model offer a promising approach for more accurate and efficient network security threat identification.