Internet Traffic Classification using MOEA and Online Refinement in Voting on Ensemble methods
Internet Traffic Classification is one of the most important issues in network management. Until now, many methods have been proposed to this end, but studies show that machine learning based algorithms have a good performance in comparison to other methods. Selecting the best feature subset causes better accuracy and efficiency in machine learning algorithms. In this paper, in order to obtain higher classification accuracy, effective features are selected using a multi-objective evolutionary algorithm. In this evolutionary algorithm, some objectives such as minimizing feature number, maximizing classification accuracy, maximizing true positive rate (maximizing TPR), and minimizing false positive rate (minimizing FPR) are satisfied simultaneously and with no conflicts with each other. In the proposed method, selected features subset is given to a new ensemble algorithm with online refinement during training and testing phases. Final result of each new ensemble algorithm is obtained by the vote of the majority with respect to the accuracy of any voter. Results show the high efficiency and performance of proposed method in comparison with other methods. So that the WWW traffic classification accuracy ascend to 99.93%. The results for 8 other traffics such as P2P indicate high accuracy of the proposed method.