IP traffic classification using machine learning

Abstract

Network traffic classification is the foundation of many network research works. In recent years, the research on traffic classification based on machine learning method is giving more accuracy. The focus of this research is to classify the traffic based on application type specially (p2p, video, http) .Performance comparison among machine learning classifiers have been done, results shows that C4.5 tree is the suitable classifier based on accuracy and stability. This offline comparison has been done by building training data sets using two methods the first use WIRESHARK for sniff the network traffic the a pre-classification has been done. The second approach is based on SNORT for sniffing and detecting the type of application and then extracted short flow by excel sheet and calculated the feature of flow Evaluate the system by using only 7 features and its response with good accuracy. The prototype system results reach average accuracy 82.7% by first data set and 88.9% by the second data set.

Finally, an idea of an online classifier has been presented based on short flows so as to quick the reaction time regarding the enhancement of the usage of the internet links