Research On Network Intrusion Detection Based On Autoencoder And Gan

With the arrival of the 5G era,virtualization technology and cloud computing continue to develop,resulting in an increasing amount of online data.However,network attacks and cybercrime are becoming increasingly common,posing a significant threat to network security.Network traffic intrusion detection is critical in ensuring the availability and reliability of network services.With the recent successful applications of deep learning in various fields,using deep neural networks to solve intrusion detection problems has become a current research focus.Currently,research on network traffic intrusion detection and classification is facing several challenges,including dynamic changes in data,imbalanced data distribution,and low accuracy of traditional intrusion detection methods.This thesis focuses on the problem of network traffic intrusion detection and its main contributions are as follows:(1)We propose an intrusion detection model,VAE-GRU-XGBoost,for classification tasks.The model combines the advantages of variational autoencoder(VAE),gated recurrent unit networks(GRU),and XGBoost models.VAE is used for feature extraction and dimensionality reduction,while GRU is used to model the time dependency in the input data.Finally,XGBoost algorithm is used for classification tasks,which can handle complex network traffic data and dynamic changing network environments.We evaluate the proposed model’s performance on the NSL-KDD dataset and compare it with the performance of using XGBoost alone.Experimental results show that the proposed model outperforms the XGBoost model in terms of classification accuracy,precision,recall,and F1 score.This demonstrates the effectiveness of our proposed model in improving XGBoost’s performance for classification tasks.(2)We propose a data augmentation algorithm based on denoising autoencoder(DAE)and Wasserstein generative adversarial network(WGAN)to generate synthetic data for the minority class in the dataset to balance the sample distribution.We use DAE for unsupervised feature extraction to capture the intrinsic structure of data;and use WGAN for generative adversarial training to enhance the generation performance of minority class samples.Experimental results show that the proposed DAE-WGAN data augmentation algorithm improves the VAE-GRU-XGBoost model’s classification accuracy,precision,recall,and F1 score on the NSL-KDD dataset.This demonstrates that the DAE-WGAN algorithm can generate high-quality minority class samples,thus achieving efficient classification discrimination.