An Unsupervised Detection-to-Mitigation Framework for Resource Exhaustion Attacks in 5G/6G Network Slicing.

Area of Science:

• Computer Science
• Network Security
• Telecommunications

Background:

• Massive Internet of Things (IoT) and sensor networks in 5G/6G systems utilize network slicing for diverse applications.
• Proportional Fair (PF) resource allocation relies on slice-reported demands, creating vulnerabilities to resource exhaustion attacks.
• Existing unsupervised defenses often lack effective mitigation strategies for detected anomalies.

Purpose of the Study:

• To propose AutoGuard-Hybrid, an unsupervised detection-to-mitigation framework for network slicing.
• To address the gap in translating anomaly detection into resource-level mitigation.
• To preserve slice-level service availability against malicious demand inflation.

Main Methods:

• Integration of Isolation Forest (IF) and Long Short-Term Memory Autoencoder (LSTM-AE) for spatial and temporal anomaly detection.
• Implementation of Adaptive Clipping and Safety Cap for allocation-aware demand purification.
• Closed-loop framework converting anomaly scores into demand purification actions before PF scheduling.

Main Results:

• AutoGuard-Hybrid demonstrates comparable performance to Isolation Forest under Continuous attacks.
• Achieved a 27.6% improvement in the mean system-wide Service Level Agreement (SLA) violation rate under Adaptive Probing attacks.
• Ablation studies show Adaptive Clipping reduced SLA violations by 75.0%, with the full pipeline achieving 84.6% reduction.

Conclusions:

• AutoGuard-Hybrid offers a practical and effective defense against resource exhaustion attacks in network slicing.
• The framework operates within the 1 ms Transmission Time Interval (TTI) constraint, suitable for next-generation IoT services.
• It successfully bridges the gap between anomaly detection and resource-level mitigation, enhancing system-wide service availability.