Construction of VAE-GRU-XGBoost intrusion detection model for network security

Area of Science:

• Cybersecurity
• Artificial Intelligence
• Machine Learning

Background:

• The increasing volume of big data amplifies network security threats.
• Complex network attacks necessitate advanced defense mechanisms.
• Deep learning offers promising solutions for network intrusion detection.

Purpose of the Study:

• To develop a robust network intrusion detection model using deep learning.
• To enhance the accuracy and efficiency of identifying network intrusions.
• To address the challenges posed by sophisticated cyberattacks.

Main Methods:

• Utilized Variational Auto-encoders (VAEs) for feature extraction and dimensionality reduction of network traffic.
• Integrated Extreme Gradient Boosting (XGBoost) for efficient classification tasks.
• Combined Gated Recurrent Units (GRUs) with VAEs and XGBoost to construct the final intrusion detection model.

Main Results:

• Achieved an Area Under the Curve (AUC) of 97.48% on the KDD99 dataset and 95.24% on the OODS dataset.
• Demonstrated classification accuracy exceeding 0.91 for various attack traffic samples in both training and testing sets.
• Reported efficient feature extraction times, ranging from 0.030s to 0.112s for sample sizes of 10,000 and 40,000.

Conclusions:

• The developed deep learning model, based on an improved Variational Auto-encoder, offers high accuracy in network intrusion detection.
• The model effectively handles complex network attacks, significantly contributing to network security.
• The integration of VAEs, XGBoost, and GRUs provides a powerful framework for real-time threat identification.