Using Ensemble Learning, A Cosine Similarity-Based Model for Detecting Security Anomalies in Software-Defined Networks

Currently, Software-Defined Network (SDN) is as one of the most commonly used network architectures. It requires using a software-based, centralized controller to communicate with the underlying network hardware to control the flow of the traffic through the network. This helps achieve easy, flexible, and integrated control and management of the entire network. In SDNs, the controller is the sole entity that monitors the entire network and is accountable for traffic management based on its comprehensive understanding of the network. Software-Defined Networks (SDNs) rely on a centralized controller for network management, yet this centralized control exposes vulnerabilities, leaving networks susceptible to disruptive attacks. To fortify SDN security, this article presents an innovative approach that merges machine learning and deep learning techniques through ensemble learning. Our method introduces several key innovations. Firstly, it employs ensemble learning, combining multiple models to enhance predictive capability. Secondly, the utilization of cosine similarity-based classification enables the grouping of attacks with varying degrees of similarity, refining the distinction between attack types. Thirdly, the approach includes customized classifiers, trained specifically for distinct attack classes, optimizing detection accuracy for each type. Finally, the K-Nearest Neighbor (KNN) algorithm is applied for final classification of new data samples, improving precision in identifying attack types. With an exceptional accuracy rate of 99.91%, our method surpasses the performance of analogous studies. By amalgamating these innovations, our approach establishes a robust framework for detecting security anomalies in SDNs, reinforcing network integrity and reliability.