You are here

Self-Learning Classifier for Internet Traffic

TitleSelf-Learning Classifier for Internet Traffic
Publication TypeConference Paper
Year of Publication2013
AuthorsGrimaudo, L., M. Mellia, E. Baralis, and R. Keralapura
Conference NameThe 5th IEEE International Traffic Monitoring and Analysis Workshop (TMA 2013)
Abstract

Network visibility is a critical part of traffic engineering, network management, and security. Recently, unsupervised algorithms have been envisioned as a viable alternative to automatically identify classes of traffic. However, the accuracy achieved so far does not allow to use them for traffic classification in practical scenario.
In this paper, we propose SeLeCT, a Self-Learning Classifier for Internet traffic. It uses unsupervised algorithms along with an adaptive learning approach to automatically let classes of traffic emerge, being identified and (easily) labeled. SeLeCT automatically groups flows into pure (or homogeneous) clusters using alternating simple clustering and filtering phases to remove outliers. SeLeCT uses an adaptive learning approach to boost its ability to spot new protocols and applications. Finally, SeLeCT also simplifies label assignment (which is still based on some manual intervention) so that proper class labels can be easily discovered.
We evaluate the performance of SeLeCT using traffic traces collected in different years from various ISPs located in 3 different continents. Our experiments show that SeLeCT achieves overall accuracy close to 98%. Unlike state-of-art classifiers, the biggest advantage of SeLeCT is its ability to help discovering new protocols and applications in an almost automated fashion.

DOI10.1109/INFCOMW.2013.6562900
Citation KeyGri2013
Project year: 
First year
WP(s) associated with the paper: 
WP3 - Large-scale data analysis
Partner(s) associated with the paper's author(s): 
Politecnico di Torino
Is this an OFFICIALLY supported mPlane paper?: 
Yes
Attachment: