An Enhanced Network Traffic Classification in Software-Defined Networking Using Softmax Regression

Abstract

Network traffic classification in Software-Defined Networking is an essential aspect of SDN and plays an important role in the efficient management and security of modern networks. The classical algorithm for traffic classification based on port numbers cannot handle the increasing encrypted traffic and dynamic port allocation. At the same time, algorithms based on machine learning, for example, flow-based and statistics-based, fail to classify accurately because of the complexity of the multi-class problem. A multi-class traffic classification framework based on the Softmax Regression algorithm is proposed to efficiently identify DNS, Telnet, Ping, and Voice flows in an SDN environment. The algorithm provides a probabilistic method for multi-class classification maitaining both interpretability and high accuracy. Normalization and parameter tuning are carried outto enhance performance of classification. The results of the experiments show that the classification accuracy is 99.68% for the Softmax regression algorithm, while it is 98.53% for the logistic regression algorithm. This illustrates the significance of Softmax regression for multi-class classification and highlights its scalability and reliability for optimizing traffic management in dynamic SDN systems.