Development of a Machine-Learning Intrusion Detection System and Testing of Its Performance Using a Generative...

Area of Science:

• Cybersecurity
• Network Security
• Machine Learning

Background:

• Intrusion detection and prevention are critical for network security infrastructure.
• Machine learning-based Intrusion Detection Systems (IDSs) are increasingly used to detect sophisticated malicious traffic.
• Attackers continuously evolve methods to evade rule-based detection systems.

Purpose of the Study:

• To demonstrate the creation of adversarial network traffic capable of evading machine learning-based IDSs.
• To implement a Generative Adversarial Network (GAN) for generating such adversarial traffic.
• To evaluate the impact of using GAN-generated adversarial traffic on IDS performance.

Main Methods:

• Utilized the NSL-KDD dataset for training and evaluating machine learning models.
• Developed a Generative Adversarial Network (GAN), a deep learning architecture, to create adversarial traffic instances.
• Tested the performance of an IDS against GAN-generated adversarial traffic.

Main Results:

• GAN-generated adversarial traffic was successfully created and demonstrated the ability to evade IDS detection.
• Testing the IDS with the generated adversarial traffic led to improved IDS performance.
• The study confirmed that adversarial traffic can be used to enhance IDS resilience against novel attacks.

Conclusions:

• Generative Adversarial Networks (GANs) offer a powerful method for creating sophisticated adversarial network traffic.
• Adversarial training using GAN-generated data can significantly improve the robustness and detection capabilities of machine learning-based IDSs.
• This approach provides a viable strategy for proactively defending against evolving cyber threats.