Publication

Abstract : Modern network infrastructures are becoming increasingly complex, and dynamic traffic patterns require new approaches for optimal topology and resource management. To address these challenges, this work proposes a hybrid framework that integrates genetic algorithms (GAs), neural networks (NNs), and the deep Q-network (DQN) model. GAs are used at two key stages: first, to optimize the network topology by tuning adaptive structures that can effectively manage the various traffic conditions; and second, to process traffic data obtained from the optimized topology to allow for prediction of future network behavior. This approach captures complex patterns in network dynamics, enabling neural networks to refine the prediction process and provide valuable input for more informed decision-making. These predictions serve as crucial input features for the DQN model, allowing dynamic resource provisioning during runtime to optimize performance under varying traffic conditions. We use the NetworkX Python library for the implementation of the system, which allows scalable and efficient simulation of complex network structures. By incorporating predictive modeling, optimization, and reinforcement learning, the system continuously adapts to changing network requirements, enhancing performance and efficiency. The results demonstrate the power of AI-driven systems in optimizing network performance. NNs provide prediction, GAs enhance optimization, and DQN models ensure adaptability. This approach offers a robust, scalable, and adaptive solution for dynamic, resource-intensive infrastructures. It marks a significant advancement in intelligent network management.