Machine Learning Based IP Traffic Classfication

TAYŞİ Z. C., Karsligil M. E., YAVUZ A. G., Sahin R., Yilmaz T., Demirel H.

21st Signal Processing and Communications Applications Conference (SIU), CYPRUS, 24 - 26 April 2013 identifier identifier

  • Publication Type: Conference Paper / Full Text
  • Doi Number: 10.1109/siu.2013.6531459
  • Country: CYPRUS
  • Yıldız Technical University Affiliated: Yes


Nowadays several topics such as improving the quality of service, bandwidth utilization, and creation of different service packages, have gained importance due to widespread use of Internet. It is crucial to identify and classify protocols and applications communicating through the network in order to perform these tasks. There are three types of systems to classify protocols and applications communicating through the network, namely, port-based, payload-based and machine learning based. In this work, we focused on Instant Messaging (IM), Peer-to-peer (P2P), Social Networks, Video and Voice-over-IP (VoIP) classes which have higher importance for the Internet Service Providers. We evaluated the performance of our system with several classifiers. Random Forest classifier has had the highest success rate among others.