Cross-Layer Federated Learning for Lightweight IoT Intrusion Detection Systems

Area of Science:

• Computer Science
• Cybersecurity
• Internet of Things (IoT)

Background:

• The rapid expansion of Internet of Things (IoT) devices presents significant security and privacy challenges.
• Securing IoT networks and their data requires efficient and privacy-preserving solutions, especially for resource-constrained devices.

Purpose of the Study:

• To propose a federated sampling and lightweight intrusion-detection system (IDS) for IoT networks.
• To enhance IoT security and data privacy using a semi-supervised, K-means-based approach for anomaly detection.

Main Methods:

• Federated learning framework for local data processing on IoT devices.
• K-means clustering for network traffic sampling and anomaly identification.
• Sharing only summary statistics to maintain data privacy.

Main Results:

• The proposed federated IDS effectively detects intrusions in IoT networks.
• The system demonstrates efficiency suitable for resource-constrained IoT devices.
• Collaboration between workers and the central coordinator can increase the true-positive rate by up to 10%.

Conclusions:

• The federated IDS offers a viable solution for IoT security and privacy.
• The system balances detection performance (precision-recall trade-offs) with privacy preservation.
• Collaborative approaches in federated learning can significantly improve intrusion detection accuracy in IoT environments.