Bayesian Networks-Based Traffic Classification Approach for Uncovering Variable Dependencies in Software-defined Edge Environment

Network traffic can provide ample information about the hidden characteristics about the actual underlying applications. Thus, it is important to uncover the hidden characteristics about he network traffic in order to further classify it for efficient packet processing. In this paper, we have opted for three different methods, a) Naive Bayes, b) Tree Augmented Naive Bayes (TAN), and c) Augmented Naive Bayes to examine their effectiveness in accurately classifying network traffic. We have used the GeNIe Modeler tool to build and analyze Bayesian networks. Further-more, we have compared the performance outcomes extracted through the use of three methods to provide valuable information about their effectiveness for network traffic classification. For this purpose, we considered network traffic dataset obtained from the University of Cambridge, covering potential applications in real-world network environments. The obtained results showcase the effectiveness of the considered methods, with Tree Augmented Naive Bayes depicting an accuracy of 95.6%, higher than the other methods.