Advanced Machine Learning Models with Metaheuristic Enhancements for Active IoT-SDN Network Traffic Analysis

Due to constant connectivity these days, it’s very important to sort and arrange the different types of network traffic well to maintain the best possible performance and computing power in IoT systems. Even so, existing ways to classify traffic in SDN may deal with a wide variety of features, poor selection of important features and a lack of accuracy during dynamic and varied IoT scenarios. Therefore, this paper presents an enhanced machine learning-based technique to optimize network traffic classification in SDN-based IoT networks. It uses traffic information from the Dalhousie NIMS Lab IoT 2024 dataset after processing it with the Bag of Words (BoW) model. PCA (Principal Component Analysis) is used to lower the number of features. With the ESO algorithm, the necessary traffic features are chosen after critical traffic is analyzed. Next, these features are improved and organized using mixed machine learning techniques, for example, Relevance Vector Machines with Secretary Bird Optimization (RVMSBO), Hybrid XGBoost enhanced by Hunter Prey Optimization (HXGBHPO) and an Adjusted Boosting-Random Forest Algorithm (ABCRFA). RVMSBO performed better than other reviewed models, with specificity of 0.98582, an accuracy of 0.98231, precision of 0.98571 and an F1-score of 0.98224. The observed results confirm that the given method helps improve both traffic classification performance and the functioning of the IoT-SDN network.