Ensemble classifier for traffic in presence of changing distributions
Runxin Wang, Lei Shi, Brendan Jennings · 2013
Traffic classification plays an important role in many short to medium term network management tasks and in long term network dimensioning/planning. In recent years a number of traffic classifiers have been proposed, in particular classifiers based on machine learning techniques exhibit high levels of accuracy. However, in practice, even if classifiers can be accurately trained at a given time, their accuracy will subsequently degrade when the characteristics of the network traffic change. In this paper, we propose an adjustable traffic classification system, the key technique of which is ensemble classification, assisted with a change detection method. Our system enables a traffic classifier to be effectively updated in response to the changing traffic distributions. Experimental results show that our classifier produces improved accuracy with relatively shorter updating time.