Enhancing IoT Network Security: Unveiling the Power of Self-Supervised Learning against DDoS Attacks

Area of Science:

• Cybersecurity
• Machine Learning
• Network Security

Background:

• The Internet of Things (IoT) is rapidly expanding, creating a vast attack surface vulnerable to data theft and denial-of-service attacks.
• Existing security measures often struggle with the scale and complexity of IoT networks, necessitating advanced detection methods.
• Manual data labeling for Intrusion Detection Systems (IDS) is resource-intensive and limits scalability.

Purpose of the Study:

• To introduce a network-based Intrusion Detection System (IDS) specifically designed for Internet of Things (IoT) networks.
• To evaluate the effectiveness of self-supervised learning against supervised learning for detecting distributed denial-of-service (DDoS) attacks in IoT environments.
• To propose an optimal training framework for contrastive learning in cybersecurity using visual representations.

Main Methods:

• Generating synthetic images from flow-level traffic data using the Bot-IoT and LATAM-DDoS-IoT datasets.
• Conducting experiments using both supervised and self-supervised learning paradigms.
• Performing extensive ablation studies to identify optimal training strategies for visual representation learning.

Main Results:

• Self-supervised learning demonstrated superior performance over supervised learning in specific classification tasks.
• The F1 score for attack detection using self-supervised learning was 4.83% higher than supervised learning.
• Multiclass protocol classification accuracy improved by 14.61% with self-supervised learning.
• Self-supervised learning transferability exceeded supervised learning by over 5% in precision and nearly 1% in F1 score in some instances.

Conclusions:

• Self-supervised learning offers a robust and efficient alternative to traditional supervised methods for IoT security.
• The proposed IDS effectively protects IoT networks against distributed denial-of-service attacks.
• Visual representation learning through contrastive methods presents a promising direction for future cybersecurity research.