A Multi-Agent System for 5G Throughput Prediction in Multi-Operator Urban Environments

This paper proposes a Tiered Multi-Agent System (TMAS) for 5G throughput prediction in heterogeneous urban environments. TMAS dynamically routes edge telemetry to context-aware Domain Micro-Agents specialized by network and usage context. The evaluation uses 48,618 samples collected in Sunway City, Malaysia, across three Tier-1 operators, three mobility modes, and three traffic profiles. The reported results reach an R² of up to 0.931 and a minimum MAE of 0.53 Mbps. Agentic routing overhead ranges from 0.004 to 0.126 ms, while the paper also reports rapid micro-agent training and low inference latency.