Design And Implementation Of Fast Network Intrusion Detection Algorithm Based On Data Dimensionality Reduction

With the widespread popularity and use of the Internet,the issue of network security has become a problem that cannot be ignored.There are endless types of network attacks,which constantly threaten the security of Internet information.The current rapid development of the Internet presents a complex and changing network environment,the large amount of network data information,high dimensionality,serious imbalance in the category and inconsistent distribution of features,etc.,which in turn increases the difficulty of network intrusion detection.In response to the above problems,this thesis conducts a study,the main contents of which are as follows.(1)To address the problem that the high-dimensional unbalanced intrusion data leads to the trained models often have low accuracy in identifying the minority class attacks,a data processing model based on the combination of Relief F and GAN is proposed.In this model,the Relief F algorithm is firstly used to screen out the features subset that can more accurately represent the unbalanced data distribution,and then the GAN sampling technique is used to sample the minority class attacks to oversampling to achieve data equalization.Finally,the data after the dimensionality reduction and equalization process is detected and classified.The experimental results show that,compared with the SMOTE sampling technique and the ADASYN sampling technique,the equilibrium state achieved by the data set processed by the GAN sampling technique can better improve the performance of the intrusion detection model.(2)Selecting the appropriate parameter combination can not only enhance the performance of the model,but also reduce the overfitting of the model in the case of data imbalance.Therefore,an optimization-based improved Light GBM intrusion detection model is proposed,and a particle swarm optimization algorithm is used to optimize the parameters of the Light GBM model to achieve fast convergence while ensuring the accuracy of the optimization search.Experimental results show that the proposed particle swarm optimization Light GBM intrusion detection model has improved accuracy and faster detection speed,and compared with grid search and genetic algorithm,etc.to verify the effectiveness of the model.(3)Experimental analysis of the proposed algorithm on two publicly available datasets,UNSW-NB15 and CIC-IDS-2017,shows that the overall performance of the proposed intrusion detection model outperforms other models.Among them,the training accuracy reaches 96.75% on the UNSW-NB15 dataset,the testing accuracy reaches93.65%,the detection time is accelerated by 6.48 s,and the accuracy of a few classes of samples is improved to some extent.(4)Designed and implemented a fast intrusion detection system based on data dimensionality reduction,and completed the requirements analysis and detailed design of the system.Through experiments and system tests,it is proved to be feasible and effective to improve the performance of intrusion detection.