A CE-GAN based approach to address data imbalance in network intrusion detection systems

Area of Science:

• Cybersecurity
• Artificial Intelligence
• Network Security

Background:

• Traditional intrusion detection systems struggle with complex network behaviors and imbalanced datasets.
• Data imbalance significantly hinders the performance and robustness of existing network intrusion detection models.
• Game theory concepts offer potential for improving classifier ensemble optimization in security applications.

Purpose of the Study:

• To propose a novel network intrusion detection system that effectively addresses data imbalance.
• To enhance the robustness and classification performance of intrusion detection systems using advanced AI techniques.
• To leverage game theory and generative adversarial networks for improved cybersecurity.

Main Methods:

• Introduced the Nash equilibrium concept from game theory into classifier ensemble optimization.
• Developed a Conditional Generative Adversarial Network with a Conditional Aggregation Encoder-Decoder Structure (CE-GAN).
• Utilized a composite loss function to ensure authenticity and diversity of generated network traffic samples.

Main Results:

• CE-GAN effectively augmented rare data samples in imbalanced datasets.
• Significant improvements in classification metrics were observed on the NSL-KDD and UNSW-NB15 datasets.
• The proposed CE-GAN model demonstrated superior performance in network intrusion detection compared to traditional methods.

Conclusions:

• The CE-GAN model provides a robust solution for network intrusion detection, particularly in scenarios with imbalanced data.
• Integrating game theory and advanced generative models enhances the capabilities of cybersecurity systems.
• This approach offers a significant advancement in detecting complex network intrusions effectively.