Optimized Traffic Classification System for Software-Defined Networking using a Deep Learning-based Approach

Abstract

Software-Defined Networking (SDN) increases scalability and flexibility of network administration by eliminating the control as well as data planes. SDN improves network management by doing away with the control and data planes, consequential in increased scalability and flexibility. A traffic classification system for SDN improves network efficiency by classifying the data flows. Quality of Services (QoS) enhances and optimizes use of resources with flexible adaptation to changing network requirements. To create an optimal traffic classification system for SDN, we proposed a novel Deep Learning (DL) approach called Lightning Search fine-tuned Generative Adversarial Networks (LS-GAN). We collected a dataset comprising several kinds of network traffic logs to train the suggested methodology. The obtained raw data is pre-processed using the Unit Vector Transformation (UVT) technique. Kernel Principal Component Analysis (K-PCA) is used with the processed data to determine the key features. The LS-GAN approach combines the potent capabilities of Generative Adversarial Networks (GANs) with blazingly quick search algorithms. The system can effectively and precisely detect different kinds of network traffic inside SDN designs by combining these methods. The proposed LS-GAN obtained a Precision (96.2%), Accuracy (98.3%), Recall (97.3%) and F1-score (98.6%). The experimental outcome show that the suggested LS-GAN approach performed better than existing approaches in SDN infrastructure for increased traffic classification.