ALINEAREMULATIONANDMACHINELEARNINGAPPROACHFOR OPTIMIZEDROUTINGALGORITHMS

Abstract

Optimizing routing decisions in networks with high traffic variability and stringent real‑time requirements is a persistent challenge. This paper proposes a hybrid approach that integrates linear emulation with machine learning to produce low‑latency, resource‑efficient routing policies. A lightweight linear emulator provides fast estimates of local network metrics (latency, packet loss, available bandwidth) while a machine learning model—configured for either supervised regression or reinforcement learning—captures global and historical traffic patterns to predict optimal next‑hop selections. The system fuses emulator outputs and ML predictions in real time to drive adaptive load distribution and route reconfiguration under changing conditions. We evaluate the approach through packet‑levelsimulationsandahardwaretestbed,comparingitagainststaticrouting schemes and baseline dynamic algorithms.