A Parallelized Network Traffic Classification Based on Hidden Markov Model

Abstract

This paper implemented a network traffic classification method on the basis of Guassian Mixture Model-Hidden Markov Model using packet-level properties in network traffic flows (PLGMM-HMM). Our model firstly builds PLGMM-HMMs via two packet-level properties, inter packet time and payload size, respectively; then, we construct the estimation function by computing the F-Measure value through classifying another training set using the PLGMM-HMMs. Hadoop Streaming based MapReduce has been evaluated while performing our classification experiment. Results show that our PLGMM-HMM based classification method could obtain considerable accuracy, giving out the accuracy over 90% on collected datasets, and comparatively outperforming classifiers based on HMMs with variables obeying other distributions. It is recommended that this framework could be applied to other machine learning methods as a multi-classifier template.