02/05/2026
We analyze three fusion strategies for stacked learning: (i) hard-label fusion, (ii) probability-level fusion, and (iii) a hybrid fusion combining both outputs. Eight traditional machine learning (ML) models are benchmarked, and the 5 top performers are integrated into the stacked ensemble. Experiments show that the proposed framework achieves 99.14% accuracy with an ultra-low inference delay of 0.0019 ms per sample, demonstrating its suitability for real-time IDS in 5G edge environments. Unlike prior IDS studies, this work provides the first comparative fusion analysis on 5GSliciNdd traffic, highlighting the benefits of feature-aware stacking for high-speed networks.
Intrusion detection in 5G networks is increasingly challenged by high-volume traffic, slice-specific behaviors, and real-time latency constraints. Existing Intr