Efficient, Accurate Internet Traffic Classification using Discretization in Naive Bayes

Accurate network traffic classification is fundamental to numerous network activities, from quality of service to providing operators with useful forecasts for long-term provisioning. In this paper, we apply the discretization method in Naive Bayes for Internet traffic identification and compare the result with that of previously applied Naive Bayes kernel estimation in AUCKLAND VI and Entry data sets. Our results show that discretization is more robust and accurate than kernel estimation. The average accuracy is improved to 97.93% and outperforms the kernel estimation by up to 4.2% in Entry data sets. For AUCKLAND VI data sets, the average accuracy is improved to 90.37% from 34.17%. We also find that discretization method for Naive Bayes is more efficient than kernel method during classification.