AI-Assisted Relay Selection Analytics for Secure Wireless Multicast Networks under Multi-Eavesdropper Conditions
Main article
Abstract
Secure wireless multicasting is becoming an important design requirement for collaborative mobile services, industrial sensor groups, emergency communication, and connected healthcare applications. Conventional physical-layer security models often study relay selection, antenna diversity, and eavesdropping risk as separate problems. This article develops an AI-assisted relay selection analytics framework for secure wireless multicast networks operating under multi-eavesdropped conditions. The framework extends the logic of partial relay selection forward strategies by using interpretable machine-learning scores to rank candidate relays according to legitimate-channel quality, eavesdropper exposure, receiver-side diversity, secrecy-rate target, and expected outage risk. Rather than proposing a purely mathematical secrecy-capacity derivation, the article emphasizes analytics design, simulation-based performance evaluation, and deployment governance. A scenario-based Monte Carlo experiment with Rayleigh fading channels compares random relay assignment, conventional partial relay selection, and AI-assisted partial relay selection across variations in average signal-to-noise ratio, receiver population, destination antenna diversity, and the number of eavesdroppers. The results indicate that AI-assisted relay selection improves the estimated probability of non-zero secrecy multicast capacity while reducing secure outage probability, especially when the relay pool is moderately large and eavesdropper density increases. The paper further discusses feature importance, latency constraints, explainability, and model-drift risks for practical multicast-security analytics. The findings show that AI-assisted relay selection is most valuable when it is used as a risk-aware decision layer that complements, rather than replaces, physical-layer security theory.
