QGuardian: A Federated Learning Framework for Quantum-Resilient Anomaly Detection in Cybersecurity

Current cybersecurity systems face two critical challenges: the impending threat of quantum computers breaking classical cryptographic algorithms, and the limitations of reactive, centralized security architectures. We propose QGuardian, a federated learning framework that combines privacy-preserving distributed machine learning, post-quantum cryptography, and blockchain-based verification for predictive anomaly detection in network security. QGuardian enables multiple distributed nodes to collaboratively train anomaly detection models without sharing raw network traffic data, while providing quantum-resistant encryption and maintaining an immutable audit trail through a blockchain ledger. We implement and evaluate the system using TensorFlow Federated for distributed learning, with post-quantum cryptographic modules (Kyber KEM) and a blockchain-based event logging system. Experimental evaluation on both synthetic data and the UNSW-NB15 real-world dataset (82,332 samples, 39 features) establishes feasibility of the integrated approach. On UNSW-NB15, federated training achieves 47.2% loss reduction over 15 rounds, with detection performance of 56.7% precision, 21.1% recall, and 30.7% F1-score. Real Kyber PQC benchmarking shows acceptable overhead (6-22× for key operations, sub-millisecond latency). The system integrates all three layers, with comprehensive evaluation frameworks for real-world datasets, Byzantine robustness (26.68× improvement shown), and performance benchmarking. This work contributes an integrated architecture and comprehensive evaluation establishing a baseline for quantumresilient, privacy-preserving cybersecurity systems that can operate in distributed environments.