A robust intelligent zero-day cyber-attack detection technique.

Area of Science:

• Cybersecurity
• Network Security
• Machine Learning

Background:

• The increasing reliance on the internet for essential services has led to a rise in cyber-attacks.
• Existing methods for detecting zero-day attacks, such as machine learning and anomaly-based approaches, have limitations in identifying low-traffic attacks and correlating network byte stream frequencies.

Purpose of the Study:

• To propose a novel, robust, and intelligent cyber-attack detection model.
• To address the limitations of current methods in detecting zero-day attacks, particularly those with low network traffic.
• To enhance the accuracy and effectiveness of zero-day attack detection.

Main Methods:

• The proposed model utilizes the concepts of heavy-hitters and graph techniques for cyber-attack detection.
• The methodology involves two phases: Signature generation and Evaluation.
• Performance is evaluated using generated signatures during the training phase.

Main Results:

• The proposed model achieved an accuracy of 91.33% for binary classification and 90.35% for multi-class classification on real-time attack data.
• On the CICIDS18 benchmark dataset, the model demonstrated a promising accuracy of 91.62% for binary-class classification.
• The results indicate superior performance compared to existing techniques, especially for detecting attacks with subtle network traffic patterns.

Conclusions:

• The developed model offers an encouraging and effective approach for detecting zero-day cyber-attacks.
• The integration of heavy-hitter and graph techniques provides a robust solution for identifying sophisticated cyber threats.
• The findings suggest a significant advancement in the field of proactive cybersecurity defense.