Publication

Abstract : Rapid growth in the number of devices connected to networks has made network analysis increasingly complex and challenging. In addition, the steady increase in network traffic volume has underscored the critical need for efficient routing to achieve quality of service (QoS). Efficient routing ensures that data flows through the network with minimal delay while optimizing resource utilization. Traditionally, algorithms like Dijkstra’s and Bellman-Ford were used for routing. However, these methods struggle in dynamic network environments characterized by traffic congestion, link failures, and bandwidth fluctuations. Reinforcement learning (RL) methods provide a solution by enabling adaptation to dynamic scenarios, allowing real-time learning and decision making. Model-free RL algorithms, such as Q-learning, determine optimal routes without requiring an explicit network model. They explore different routes, evaluate their costs using a reward-based approach, and iteratively learn the most efficient routes. This paper explores Deep Q-Learning (DQL) for network routing, which uses neural networks to approximate Q values for enhanced performance. The DQL approach is evaluated in a large-scale network scenario, demonstrating superior routing efficiency compared to the traditional Q-learning method.