Network Traffic Classification Based On A Multi-Objective Adaptive Evolutionary Algorithm

With the steady development of the Internet,the number of Internet users is increasing,and new network services are constantly emerging.Network multimedia services(online video services,web browsing,online music,etc.)as a commonly used Internet business,occupy a large part of network traffic.Therefore,it is of great significance to identify network traffic efficiently and accurately,and it is helpful for ISP(Internet Service Providers)to rationally allocate resources,manage networks and ensure quality of services.Based on the information gain ratio and evolutionary algorithm,a new multi-objective adaptive evolution feature selection algorithm,GR-MAEA,is proposed.Firstly,the method utilizes the information gain ratio for feature ranking,quickly reduces dimension,and selects the top-ranked features as the initial population of the multi-objective evolutionary algorithm.Then,the traditional evolutionary algorithm is optimized,and the inconsistency and the dimension of feature subset are selected as the objective function,which reduces the time overhead of feature selection iteration and improves the performance of classifier.Experiments show that the proposed method has higher classification accuracy and lower time complexity than existing feature selection methods on different data sets.GR-MAEA is used to select distinguishing feature combinations to identify six kinds of network traffic.In order to achieve fine-grained classification,a multi-layer K-nearest neighbor(KNN)classifier cascade scheme is designed in this thesis,and the statistical features selected by the feature selection algorithm are used to identify the network traffic and the classification results are compared with the existing multi-layer Support Vector Machine(SVM)classification scheme.The experiment shows that our method has better classification performance.Try to use a deep learning method-deep neural network(DNN)to classify the network traffic.In the experiment,a three-layer DNN is adopted to automatically select and classify the traffic,and the classification accuracy is optimized by adjusting the number of hidden layer nodes and the number of training iterations.Six types of network applications are identified effectively.The classification results are compared with those by the multi-layer SVM and multi-layer KNN.