CLEVER Project advanced autonomous, AI-driven security for 6G networks, demonstrating how deep learning and unsupervised methods outperform traditional approaches:
AI for DoS Detection & Mitigation
- Attack-Specific Adaptation: Achieved 97.3% F1-score for Hulk and 92.2% for Goldeneye using deep embedding clustering, distinguishing unique request patterns and HTTP Keep-Alive abuse.
- Unsupervised Advantage: Autoencoder + Gaussian Mixture Models (GMM) detected threats without labeled data, maintaining 92%+ accuracy on zero-day variants—unlike supervised models (54.46% drop when tested on unseen attacks).
Real-Time, Explainable Security
- Latent Space Insights: t-SNE visualizations revealed how AI separates normal (red) from malicious (blue) traffic in 16D latent space.
- Base Station Autonomy: Localized AI at 6G base stations enables real-time traffic inspection, mitigating threats before core network disruption.
- Optimized Efficiency: Packet flow length N=10-20 balanced 97% F1-scores with computational load (30M+ MACs at N=50).
Beyond Traditional Defenses
- Federated Learning Potential: Collaborative AI across base stations could revolutionize distributed threat detection.
- Proactive Cost Savings: AI embedded in base stations minimizes $5,600/minute downtime risks by identifying attacks as they emerge.
🔗 Explore the research:
CLEVER Project: https://www.cleverproject.eu/
Full Paper: https://zenodo.org/records/11045714