SecureTrust-FL: trust-aware privacy-preserving federated learning for network intrusion detection

Area of Science:

• Cybersecurity
• Machine Learning
• Distributed Systems

Background:

• The rise of IoT and distributed networks necessitates privacy-preserving intrusion detection systems.
• Existing systems struggle with heterogeneous and non-IID data conditions.
• Federated Learning (FL) offers a decentralized approach but requires robust security and trust mechanisms.

Purpose of the Study:

• To propose SecureTrust-FL, a novel trust-aware federated learning framework for privacy-preserving intrusion detection.
• To integrate Federated Learning, Blockchain, Differential Privacy, Adversarial Learning, and Zero-Trust principles.
• To enable secure collaborative learning without raw data sharing.

Main Methods:

• Federated Learning for distributed model training.
• Blockchain for immutable trust management and accountability.
• Differential Privacy to protect individual data contributions.
• FGSM-based Adversarial Learning for robustness evaluation.
• Zero-Trust Security principles for enhanced security posture.

Main Results:

• Achieved high performance across benchmark datasets (CICIDS2017, UNSW-NB15, BoT-IoT) with Accuracy (92.91%), Balanced Accuracy (93.25%), Macro F1-Score (92.89%), and AUC-ROC (95.50%).
• Demonstrated effective learning from heterogeneous, non-IID data while preserving privacy.
• Highlighted the impact of class imbalance on performance, emphasizing Balanced Accuracy and F1-Score.
• Showcased the privacy-utility trade-off with Differential Privacy and the need for enhanced adversarial defense.

Conclusions:

• SecureTrust-FL provides an effective architecture for privacy-preserving collaborative intrusion detection.
• The framework successfully integrates trust management, privacy protection, and secure federated learning.
• Blockchain-based trust ledger enhances transparency and accountability in collaborative learning.