Uncovering Relations between Traffic Classifiers and Anomaly Detectors via Graph Theory

Network traffic classification and anomaly detection have received much attention in the last few years. However, due to the the lack of common ground truth, proposed methods are evaluated through diverse processes that are usually neither comparable nor reproducible. Our final goal is to provide a common dataset with associated ground truth resulting from the cross-validation of various algorithms. This paper deals with one of the substantial issues faced in achieving this ambitious goal: relating outputs from various algorithms. We propose a general methodology based on graph theory that relates outputs from diverse algorithms by taking into account all reported information. We validate our method by comparing results of two anomaly detectors which report; traffic at different granularities. The proposed method succesfully identified similarities between the outputs of the two anomaly detectors although they report distinct features of the traffic.