Graph-Based Deep Learning Models for Optimized Routing in Autonomous Networks Using Reinforcement Learning

The complexity of modern networks is increasing, and new sophisticated routing methods are needed in the face of dynamic situations. Traditional routing algorithms are highly used, but they are subjected to limitations in handling real-time changes and scalability in a large complex network. Such techniques tend to fail in finding an optimal latency, throughput, and packet delivery ratio in dynamic environments where conditions change over time. This study proposes a new methodology known as GNN-RL, or Graph Neural Networks combined with Reinforcement Learning, to overcome traditional shortfalls by applying the graph-based structure of networks and adaptive decision-making capabilities of RL in dynamically modifying routing decisions, thus optimizing network performance in real time. Experimental results show that GNN-RL outperformed the traditional algorithms: achieving superior throughput (40%), latency (25%), and packet delivery ratio (20%). It resulted in a 90% improvement in overall performance when compared to Dijkstra's Algorithm, AODV, and OSPF at 80%, 80%, and 75%, respectively. The model is well set to adapt and optimize its routing decisions in real-time environments; it is promising for autonomous management of future networks. Results confirm that GNN-RL could offer improved routing strategies with the power to be efficient, scalable, and robust in comparison to traditional routing techniques.