Research On Encryption Traffic Classification Method Based On Deep Learning

With the development of internet technology and people’s growing concerns about privacy and security,various encryption protocols have emerged in the Internet.The widespread use of encrypted traffic makes it difficult for traditional network security measures to effectively detect and defend against it,making research on encryption traffic classification technology an important approach to solving network security problems.The increase in encrypted traffic also affects network management and maintenance,and studying encryption traffic classification technology helps address network management issues.With the continuous development of artificial intelligence technology,deep learning techniques have been widely applied in the field of encryption traffic classification to improve the accuracy and efficiency of classification.The main work and innovations of the paper are as follows:1.In the field of encrypted traffic research,collecting labeled encrypted traffic datasets is challenging,while collecting unlabeled datasets is relatively easier.This paper proposes a semi-supervised encryption traffic classification model called BCSGAN based on Deep Convolutional Generative Adversarial Networks(DCGAN).Specifically,the CNN-based network in the discriminator of DCGAN is improved by cascading BiLSTM and CNN networks to effectively extract the spatiotemporal features of the traffic and enhance the classification performance.Experimental results demonstrate that,on the same dataset,BCSGAN outperforms other traffic classification models.It achieves a classification accuracy of 94.02% on a labeled dataset of 10%,indicating its strong ability in recognizing encrypted traffic.2.Due to differences in network application usage and other factors,there exists a class imbalance issue in encrypted traffic datasets.To address this problem,this paper proposes a data augmentation model called CDCGAN based on Generative Adversarial Networks(GAN)specifically designed for encrypted traffic.The model utilizes traffic labels as additional conditions to guide the GAN network in generating specific types of small-scale traffic samples.Leveraging the advantages of DCGAN in image generation,the model transforms one-dimensional traffic into two-dimensional grayscale images and performs data augmentation through image generation techniques.Experimental results demonstrate that,in the classification experiments based on BCSGAN,the accuracy of the dataset is enhanced from 94.15% to 96.98% after data augmentation.Compared to classic data augmentation methods such as oversampling and SMOTE,CDCGAN achieves an improvement of approximately 3% in classification accuracy,providing a new approach to alleviate class imbalance in datasets.3.In the research of encrypted traffic classification,there are two approaches:statistical feature-based and raw data spatio-temporal feature-based.The former lacks effective general statistical features when facing different types of classification tasks,while the latter may disrupt the statistical features of the data during data preprocessing.To address these issues,this paper proposes a fusion-based encrypted traffic classification network called FEAFU,which combines statistical features and raw traffic spatio-temporal features.The FEAFU model utilizes autoencoders for dimensionality reduction of statistical features and residual networks for extracting spatio-temporal features from encrypted traffic.The two types of features are then fused to obtain comprehensive features for the classification task.Experimental results demonstrate that the model achieves a classification accuracy of 99.79%,outperforming other classical traffic classification models.