Improving Software-Defined Network Efficiency With Random Forest Model

Abstract

The increasing demand for real-time services, data-intensive applications, and scalable network infrastructures has challenged traditional networking models, highlighting the need for intelligent, adaptive solutions. Software-Defined Networking (SDN) addresses this by separating the control and data planes, allowing for centralized and programmable network management. However, existing SDN systems still struggle with issues such as latency, controller bottlenecks, and inefficient traffic handling under dynamic conditions. This study presents a novel approach to enhancing SDN efficiency by integrating a Bayesian-optimized Random Forest (RF) model within the controller framework. The proposed methodology involves detailed data preprocessing, feature selection using Recursive Feature Elimination  (RFE) and Principal Component Analysis (PCA), and model training with hyperparameter tuning via Bayesian optimization. Real-time integration of the RF model enables intelligent flow classification and decision-making within the SDN controller. Experimental results indicate a dramatic improvement in throughput, lower latency, and the total removal of packet loss compared to the baseline. The model reached a maximum accuracy of 99.99%, outperforming existing solutions and revealing the potential of ensemble learning in contemporary network environments. The study contributes to a solid and practical approach to designing intelligent, data-driven SDN systems.