An Improved Classification Model for Wide Area Networks with Low Speed Links (Only Abstract)

Abstract

The task of network administrators to identify and determine the type of traffic traversing through the network is very critical with the rapid growth of new traffic each day. Considering wide area networks with limited resources in terms of low speed links, quantified amount of packets are likely to be lost which lowers the quality of service. The classification procedure in such scenarios can also be affected due to the limited features extracted from the various fragments of packets that will successfully get to the destination node or server. We propose a hybrid cluster and label algorithm, which is able to classify application traffic or packets, utilizing restricted traffic features, few packets and at the same time maintains a low complexity and good classification accuracy. A wide area network exposed to extreme packet loss scenario is designed and implemented using OMNET ++ simulation to generate a dataset. The proposed model is built and tested in MATLAB simulation environment. Evaluation results shows that our proposed semi-supervised algorithm achieves an accuracy of 92.4% in classification with lower error rates of 7.4% and 2.9839 seconds processing time.