Network Traffic Classification Based on Multi-Classifier Selective Ensemble

Network traffic classification is one of the foundations of recognizing, managing, and optimizing various network resources. Ensemble learning can be used to improve the classification accuracy. A study of the method shows that selecting part of the base classifiers to do ensemble learning can get better generalization ability than selecting all, as the number of the base classifiers is large. This paper proposes a novel network traffic classification method based on multi-classifier selective ensemble (NTCMCSE) and diversity measures, which chooses the classifiers with higher accuracy from the generated ones, and then selects the diverse classifiers with improved disagreement measure strategy to do ensemble prediction. It is shown that, compared with Bagging and GASEN, NTCMCSE achieves better generalization ability, less selection time of classifiers and fewer selected base classifiers.