This paper presents a comprehensive review and experimental analysis of machine learning algorithms applied to load balancing in Software-Defined Networking (SDN). From a review standpoint, it considers the rapid development of data centers and the increasing complexity of traffic, underscoring the shortcomings of traditional algorithms. The studies examined are systematically compared in terms of their pros and cons in real-life situations. In addition to the review, the paper presents original findings by assessing specific machine learning techniques, notably Artificial Neural Networks (ANN) and Deep Learning (DL) models, via tenfold cross-validation. The experimental results indicate that the ANN achieves the lowest average response time (1.955 ms), closely followed by the DL (1.962 ms), which demonstrates the highest stability across runs. SVM with C=100C=100C=100 gets the best classification accuracy in ten-fold cross-validation. DL comes in third, and ANN comes in last. When considering both latency and accuracy, DL offers the best overall trade-off between speed, stability, and accuracy for SDN load balancing. It beats traditional baselines (LR: 4.461 ms; SVM: 17.9–18.0 ms). These results, which are directly related to the method used, show that advanced ML methods are better for SDN load balancing. The paper also addresses significant challenges, such as scalability and adaptability, and proposes future research directions for hybrid AI-driven models to enhance the efficiency of SDN-based network management.
ahmed hadi (Fri,) studied this question.