Internet traffic classification using SVM with flexible feature space

SVM is a typical machine learning algorithm with prefect generalization capacity, which is suitable for the internet traffic classification. At present, there are two approaches, One-Against-All and One-Against-One, proposed for extending SVM to multi-class problem like traffic classification. However, these approaches are both based on a unique feature space. In fact, the separating capacity of a special traffic feature is not similar to different applications. Hence, flexible feature space for extending SVM was proposed, which constructs independent feature space with optimal discriminability for each binary-SVM and trains them under their own feature space. Finally,these trained binary-SVM were ensemble by One-Against-All and One-Against-One approaches. The experiments show that the proposed approach can efficiently improve the precision and callback of the traffic classifier and easily obtain more reasonable optimal separating hyper-plane.