Network Intrusion Dataset Assessment

Research into classification using Anomaly Detection (AD) within the field of Network Intrusion Detection (NID), or Network Intrusion Anomaly Detection (NIAD), is common, but operational use of the classifiers discovered by research is not. One reason for the lack of operational use is most published testing of AD methods uses artificial datasets: making it difficult to determine how well published results apply to other datasets and the networks they represent. This research develops a method to predict the accuracy of an AD-based classifier when applied to a new dataset, based on the di erence between an already classified dataset and the new dataset. The resulting method does not accurately predict classifier accuracy, but does allow some information to be gained regarding the possible range of accuracy. Further refinement of this method could allow rapid operational application of new techniques within the NIAD field, and quick selection of the classifier(s) that will be most accurate for the network. The original document contains color images.