فهرست نشریات علمی دسترسی آزاد ایران

Software-defined networking (SDN) separates the control and data planes that utilizes a centralized controller for efficient management and control. By collecting traffic flow attributes through the southbound interface, SDN enables flexible packet forwarding. However, high traffic volumes in large-scale dynamic networks, such as data centers, can cause congestion, packet loss, and delays. Real-time traffic forecasting is essential to mitigate these issues but it requires capturing complex spatiotemporal relationships which traditional models struggle to achieve. This paper addresses these challenges by extending an SDN dataset using traffic traces from a real-world network and introducing a novel hybrid traffic prediction model that combines long short-term memory (LSTM) and gradient boost regression (GBR) for accurate traffic prediction in SDN environments. This study creates a dataset with complex models for classification and regression. The proposed model predicts network throughput based on four critical routing metrics: hop count, latency, packet loss, and queue length. The predictions are utilized within a smart selection protocol, namely the adaptive selection protocol (ASP), to dynamically adjust routing decisions to significantly enhance the overall network performance. Simulation results demonstrate that the proposed LSTM-GBR model achieves an accuracy score of 0.996 for predicting throughput values while reducing congestion and latency and improving throughput by 39% and 200% when integrating the prediction process with the ASP technique compared to the traditional prediction methods. Implemented using Mininet with Python, the proposed approach demonstrates significant improvement in network efficiency and scalability, offering robust solutions for real-time traffic management in SDN environments.