Research On Intrusion Detection Based On Deep Learning

The wave of digitalization is coming.Now,it is integrated into the development process of various industries.Although the Internet has brought convenience to life,it has also provided growing soil for endless cyberattacks.Therefore,the importance of intrusion detection has risen accordingly.In recent years,the emergence of various detection methods has opened up some new ideas for intrusion detection.First of all,this paper collected and analyzed the advanced cases in the field of intrusion detection at home and abroad.Secondly,its advantages and disadvantages have been analyzed.Finally,it targeted the use of deep learning methods for intrusion detection research.The core work of this paper is listed as follows:(1)Due to the superiority of neural network in the field of intrusion detection,this paper reproduced a variety of artificial neural network models,which can be applied in the field of intrusion detection.Then,it carried out related experiments and comparisons.Therefore,it concluded that the problem of data dimensionality reduction is solidified of intrusion detection classification models.In the process of studying and using the architecture of the neural network,it found that the traditional loss function has the problem of unbalanced distribution of a large number of attack data categories.(2)Aiming at the problem of curing the current data dimensionality reduction method,it can be combined with the convolutional neural network by introducing an improved support vector machine recursive feature elimination method.A convolutional neural network algorithm based on RFE+SVM dimensionality reduction is proposed.Then,the intrusion detection data classification model based on convolutional neural network is constructed.The model used the KDD99 data set for experiments,compared and analyzed the results.It improved the effective discrimination rate for abnormal data,and discovered some unknown attack types.(3)Aiming at the loss offset problem of the current deep learning model predicting the loss function of the classification layer which is affected by the size and complexity of the data.Based on the optimization scheme of the loss function and combined with the gradient coordination mechanism,a RFE+SVM-FLC convolutional neural network algorithm for updating the adjustment factor is proposed.By using the NSL-KDD dataset for model training,it was analyzed and compared with traditional machine learning or neural network algorithms such as BP,SVM,CNN,DBN.The model was found to perform better when dealing with data imbalanced set classification tasks.Based on the field of intrusion detection,this paper collected relevant deep learning algorithms.Aiming at the related problems of existing deep learning networks,it proposes two classification models with different emphasis.Then,after the analysis and judgment of the follow-up experiments,the shortcomings were found.Finally,it may play a certain reference for the performance improvement of the subsequent classification models in the field of intrusion detection.