Research On Network Intrusion Detection Based On Deep Learning

With the increasing penetration of the Internet into people's normal lives,types and quantities of network attacks have also increased.Identifying attack behaviors accurately and quickly becomes an essential research issue.Network intrusion detection systems are able to identify attack behaviors by analyzing the features of network traffic.Aiming at the problem of imbalanced network traffic distribution,this thesis proposes an improved algorithm for synthesizing minority samples which is suitable for network traffic data.In order to optimize the detection performance of the intrusion detection model,this thesis proposes a network intrusion detection model based on ensemble deep learning.The main work includes the following two aspects:(1)An improved Categorial Synthetic Minority Over-sampling Technique(C-SMOTE)is proposed.The algorithm first distinguishes quantitative and categorical features of minority samples.Then quantitative features are synthesized using the SMOTE algorithm,and categorical features are synthesized based on the categorical features of their k nearest neighbors.The newly synthesized features through C-SMOTE algorithm are closer to the original network traffic features,thereby improving the accuracy of the network traffic intrusion detection model.Simulation results show that based on the same intrusion detection model,the overall detection accuracy of C-SMOTE algorithm is 3.27%higher than SMOTE algorithm,and the recall rate of U2R attacks is increased by 28.50%.Compared with the cost function algorithm,C-SMOTE algorithm improves the overall accuracy rate by 3.47%and the recall rate of U2R attacks by 16.75%.Simulation results show that C-SMOTE algorithm effectively improves the overall accuracy of the network traffic intrusion detection system and the recall rate of minority attacks.(2)An ensemble network traffic intrusion detection model based on deep learning is proposed.The ensemble model integrates convolutional neural networks as well as long-term and short-term memory networks according to the stacking algorithm.K-fold cross-validation is applied in the model to alleviate the problem of overfitting.Simulation results show that the detection accuracy of the ensemble model is 4.15%higher than that of a single classifier-based intrusion detection model.In particular,this thesis combines the C-SMOTE algorithm with the ensemble deep neural network-based network traffic intrusion detection model to build a complete wireless network intrusion detection system.The overall detection accuracy of this system is 1.96%higher than that of deep learning-based intrusion detection systems,and it is 2.05%higher than other intrusion detection systems based on ensemble learning.In summary,this research aims to explore the effectiveness of data balancing and ensemble deep neural network-based network intrusion detection system.Through experimental simulation,it is verified that the proposed model has improved both detection accuracy and attack recall rate.