Evaluation Scheme for Traffic Classification Systems

Abstract

Recent research on evaluation and comparison of traffic classification systems only used tagged offline dataset, thus the result can only reflect the performance of the classification systems on the network from which the offline dataset was collected. Besides, the difference of scopes and granularities of different traffic classification systems also render them not comparable. In this work, we propose a novel two-phased evaluation system which combines offline dataset evaluation and online evaluation. Our evaluation approach can help network manager pick the traffic classification system that fit their specific network most. In addition, we introduce three metrics corresponding to our evaluation scheme to do comprehensive evaluation and group applications according to their behaviors and functions to compare classification systems of different granularities.