Human-guided auto-labeling for network traffic data: The GELM approach.

Area of Science:

• Computer Science
• Cybersecurity
• Machine Learning

Background:

• Accurate data labeling is essential for supervised learning and classification in network security.
• Existing methods may lack efficiency and consistency in labeling large datasets.

Purpose of the Study:

• To develop a human-guided auto-labeling algorithm using self-supervised learning for fast, accurate, and consistent data labeling.
• To evaluate the algorithm's performance on network traffic datasets, including various cyberattacks.

Main Methods:

• A three-process algorithm: auto-labeling with weighted features, validation using Generalized Extreme Learning Machine (GELM), and an update process for new data.
• Application to five network traffic datasets, including Distributed Denial of Service (DDoS), DoS, BruteForce, and PortScan attacks.

Main Results:

• The algorithm demonstrated quick, accurate, and consistent labeling of unlabeled datasets.
• Generalized Extreme Learning Machine (GELM) enabled real-time data labeling.
• High performance similarity between auto-labels and conventional labels was observed on DDoS-only datasets.
• Performance differences were noted on datasets with diverse attack types, suggesting areas for further investigation.

Conclusions:

• The human-guided auto-labeling algorithm offers a viable criterion for real-time data labeling in various applications, especially network security.
• Further research is needed to address performance variations on complex, multi-attack datasets.