Network Traffic Classification Using Machine Learning Algorithms

With beginning of extensive usage of internet on our daily lives, to control network flows and manage network traffic for efficient usage of network sources has become more and more important. Traditional query and reporting tools started to become insufficient, so, researches done to decide how these data will be used has made machine learning and data mining concepts to have much more value in our lives. Traditional approaches performed for network analysis (port-based, payload-based) has become insufficient with the changes on network usage behaviors and technology. Therefore, Machine Learning based network traffic classification technique has been developed.On this thesis study, by the usage of network traffic data set, attribute set for network flows will be calculated, and C4.5, Naive Bayes, EM and K-means algorithms which are used frequently on classification and clustering analysis will be tested using comparison metrics. Also, effects on classification algorithms of Adaboost, used for boosting algorithms performances will be tested. Most convenient attribute set will be determined by using attribute selection algorithms which helps to be removed unused and unnecessary attributes on generated attribute set. Effects of this attribute set on algorithms' performance will be researched.