Data augmentation-based conditional Wasserstein generative adversarial network-gradient penalty for XSS attack...

Area of Science:

• Cybersecurity
• Machine Learning
• Web Application Security

Background:

• Web applications face increasing cyberattacks, particularly cross-site scripting (XSS).
• Existing machine learning (ML) models struggle with unbalanced XSS datasets and diverse attack vectors.
• Limited data hinders the efficiency of ML-based XSS detection systems.

Purpose of the Study:

• To enhance XSS detection systems in low-resource environments.
• To address the challenge of limited and unbalanced training data for ML algorithms.
• To improve the robustness and accuracy of cybersecurity defenses against XSS attacks.

Main Methods:

• Proposed a conditional Wasserstein generative adversarial network with gradient penalty (cWGAN-GP).
• Integrated conditional GAN and WGAN-GP for data augmentation from limited, unbalanced datasets.
• Generated synthetic minority class samples representative of real XSS attack scenarios.

Main Results:

• The proposed model successfully generated valid and reliable synthetic XSS attack samples.
• Augmented data significantly improved the performance of a boosting model for XSS detection.
• Experimental results demonstrated superior performance compared to state-of-the-art methods on unbalanced datasets.

Conclusions:

• The cWGAN-GP effectively addresses data scarcity and imbalance issues in XSS detection.
• The generated synthetic data enhances the strength and reliability of cybersecurity systems.
• This method offers a promising solution for improving web application security against sophisticated XSS attacks.