Comparative analysis of five machine learning algorithms for IP traffic classification

Abstract

With rapid increase in internet traffic over last few years due to the use of variety of internet applications, the area of IP traffic classification becomes very significant from the point of view of various internet service providers and other governmental and private organizations. Now days, traditional IP traffic classification techniques such as port number based and payload based direct packet inspection techniques are seldom used because of use of dynamic port number instead of well-known port number in packet headers and various encryption techniques which inhibit inspection of packet payload. Current trends are use of machine learning (ML) techniques for this classification. In this research paper, real time internet traffic dataset has been developed using packet capturing tool and then using attribute selection algorithms, a reduced feature dataset has been developed. After that, five ML algorithms MLP, RBF, C4.5, Bayes Net and Naive Bayes are used for IP traffic classification with these datasets. This experimental analysis shows that Bayes Net and C4.5 are effective ML techniques for IP traffic classification with accuracy in the range of 94 %.