Comparing the Performance of Machine Learning Algorithms for Multiclass Network Traffic Classification

Abstract

As we move into a technology-based generation, more devices will be in use, increasing network traffic activity. High levels of activity force network operators to be able to segregate traffic quickly and efficiently. This issue brings light to the importance of network traffic classification. Network traffic classifiers are expected to be able to identify the network service that a particular traffic flow belongs to, which is crucial information for network management and security. Recently, the research area has turned to machine learning (ML) algorithms to act as classifiers. With just a few attributes, machine learning classifiers can group traffic flows into classes with high accuracy. In this study, our objective was to compare various machine learning algorithms and find the optimal algorithm for network traffic classification. We used a dataset that consisted of 100,000 traffic flows and 40 distinct network applications. We applied five machine learning algorithms: Artificial Neural Network, Decision Tree, Naive Bayes Classifier, Random Forest, and Support Vector Machine. 

Our results showed that the Random Forest and Decision Tree classifiers were the most accurate models, suggesting further use of these algorithms in the future of network traffic classification.