Impact of Adversarial Learning on XGBoost and Random Forest Using Tabular Data

Ensuring the robustness of machine learning models has become critical for their safe application in organizational settings. In the field of cybersecurity, weak models can lead to severe consequences for organizations. A model that misclassifies network traffic containing malware as benign becomes unusable. This research investigates the resilience of two popular models—XGBoost and Random Forest—under adversarial conditions in the context of cybersecurity. The findings reveal that both Random Forest and XGBoost experienced significant performance degradation to unacceptable levels in a black-box scenario when exposed to adversarial data. Despite the application of hyperparameter tuning and feature engineering, the decline in performance remained substantial.

Author

• Koonjbearry, Rajin

Language

• English

Keyword

• Tabular Data
• Zeroth Order Optimization
• Ensemble Learning
• Random Forest
• XGBoost
• Adversarial Learning

Date created

• 2024

Type of Work

• Dissertation

License

• All rights reserved

Rights statement

• In Copyright

GW Unit

• Cybersecurity Analytics

Degree

• D.Engr.

Advisor

• Buddha, Nepal

Committee Member(s)

• Etemadi, Amir
• Morton, John

Persistent URL

•  https://scholarspace.library.gwu.edu/etd/5m60qs80b