# The Threat of Adversarial Attacks on Machine Learning in Network Security -- A Survey

**Authors:** Olakunle Ibitoye, Rana Abou-Khamis, Ashraf Matrawy, M. Omair Shafiq  
**arXiv:** [1911.02621](https://arxiv.org/abs/1911.02621)  
**Date:** 2019-11-06  
**Summary page:** https://www.summarizepaper.com/en/paper/1911.02621v1/

## Key points

- Machine learning models have revolutionized decision support systems
- Machine learning in network security faces a disproportionate threat of active adversarial attacks
- Adversaries constantly probe machine learning systems with inputs designed to bypass the system and induce incorrect predictions
- The survey provides a comprehensive analysis of the threat posed by adversarial attacks on machine learning in network security
- Taxonomy of machine learning techniques styles and algorithms is presented
- Classification of machine learning in network security applications is introduced
- Various types of adversarial attacks against machine learning in network security are examined
- Two classification approaches for adversarial attacks in network security are introduced: taxonomy-based and problem space vs feature space dimensional model-based
- Different defenses against adversarial attacks on machine learning-based network security applications are analyzed
- An adversarial risk model is introduced and several existing adversarial attacks are evaluated using this model
- The survey identifies where each attack classification resides within the adversarial risk model, providing insights into the threat landscape faced by machine learning models deployed in network security applications

## Summary

Machine learning models have revolutionized decision support systems by making them faster, more accurate, and more efficient. However, when it comes to network security applications, machine learning faces a disproportionate threat of active adversarial attacks compared to other domains. This is because machine learning applications in network security, such as malware detection, intrusion detection and spam filtering are inherently adversarial in nature. In an ongoing arms race between attackers and defenders adversaries constantly probe machine learning systems with inputs specifically designed to bypass the system and induce incorrect predictions. To address this issue this survey provides a comprehensive analysis of the threat posed by adversarial attacks on machine learning in network security. The survey begins by presenting a taxonomy of machine learning techniques styles and algorithms. It then introduces a classification of machine learning in network security applications. Next various types of adversarial attacks against machine learning in network security are examined. Two classification approaches for adversarial attacks in network security are introduced: one based on a taxonomy of network security applications and another based on a problem space vs feature space dimensional classification model. Furthermore the survey analyzes different defenses against adversarial attacks on machine learning-based network security applications. It concludes by introducing an adversarial risk model and evaluating several existing adversarial attacks using this model. The survey also identifies where each attack classification resides within the adversarial risk model providing valuable insights into the threat landscape faced by machine learning models deployed in network security applications and offering a framework for understanding and mitigating these threats effectively.