Research On Encrypted Traffic Classification Technology Based On SOM-K Fusion Algorithm

With the explosive growth of network traffic,and the increased requirement of privacy protection,the traditional port-based and payload-based traffic classification methods are no longer satisfied the fast real-time classification.In the Internet era,network traffic classification technology plays a vital role in improving network management efficiency,making a better experience,and ensuring a safe network environment.Therefore,this article will research traffic classification on the following three aspects.First of all,in the face of the difficulty of extracting effective information from encrypted traffic,thesis proposed a method for generating encrypted traffic data based on improved DCGAN.This method constructs the traffic feature matrix as the training set of the DCGAN model that shows the pixel-like behavior as an image matrix.To learn the timing characteristics of flow data better,adding an LSTM layer in the traditional DCGAN model.Experiment results show that the improved DCGAN model can learn the characteristics of the input data better,and the training time is shorter than traditional DCGAN model.Then,in order to improve the classification accuracy of unlabeled traffic data and improve the shortcomings of the K-means algorithm,thesis proposes a network traffic classification model based on the SOM network and K-means fusion algorithm.The stage of SOM algorithm is used to solve the instability caused by the randomness of the K-means algorithm.The experiment compares the performance of the existing improved K-means algorithm with the existing improved K-means algorithm in terms of classification accuracy and accuracy.The experimental results show that there is a certain improvement in accuracy and stability.Finally,in order to evaluate the clustering results more reasonably,thesis implements an encrypted traffic classification model that based on time series characteristics and used the S?Dbw coefficient as the evaluation index of the clustering results.First,select a small amount of labeled traffic data to form a pseudo-image matrix as the input of the improved DCGAN discriminator,then use the training results of DCGAN as the training data set of the SOM-K fusion algorithm,and finally use the S?Dbw coefficient as the evaluation index of the clustering results.Experiments show that the robustness of the S?Dbw coefficient is better than other clustering result evaluation indicators.The classification model proposed in thesis is better than other improved K-means and SOM in terms of classification accuracy,precision and S?Dbw coefficient.