Analysis of Datasets for Network Traffic Classification

Abstract

Support vector machines play a major role in network traffic classification. These Machine Learning techniques use labeled datasets for training and cross validation purposes. They require mathematical and statistical properties that truly represent the characteristics of the network traffic in order to perform efficient classification tasks. In this paper we present a feature extraction learning technique and a distance metric learning approach to support the classification of http and https traffic, based on the findings that the relationship between the congestion window size and the packet length can be characterized by overlapping rectangular geometric patterns. We demonstrate that the support vector machines with these learning approaches can give desirable classification results than one without these representation learning approaches.