Research On Intrusion Detection Based On Deep Learning

Nowadays,it has been very common for people to use computer,so network security problem has become a focus gradually.With the increasing of network attacks frequently and diversified,the requirements for network security are getting higher and higher.Intrusion detection is an important network security technology,which can discover the network security risks and determine the type of attack immediately,so as to intercept the attacks in time and protect the network from infringement.Traditional intrusion detection technology mainly relies on experts' experience,it is difficult to respond to new types of attacks,resulting in the limitations of intrusion detection technology.At the same time,there is a problem of unbalanced data in intrusion detection,which makes the results biased and affect the model's classification performance.In this paper,the intrusion detection technology is studied based on the deep learning method,This paper studies intrusion detection technology based on deep learning methods.Using deep learning has the advantage of extracting main features from complex features,so that intrusion detection models can quickly learn the main features of network attacks and accurately classify them.The main research contents and innovations of this article are as follows:(1)A classification model t-SNE-LSTM(t-distributed stochastic neighbor embedding LSTM)based on a combination of nonlinear dimensionality reduction method and long-short-term memory network is proposed,which can better perform nonlinear invasion sequences classification.The model uses the t-SNE algorithm to process many high-dimensional non-linear raw data to obtain low-dimensional data;and then uses bidirectional LSTM as a training network for low-dimensional data to obtain classification results.Because LSTM is good at processing time series data,the model gets better results.(2)A model SMOTEAdaboost.M2 is proposed to deal with data imbalance,which can effectively improve the classification effect of a few types of data in intrusion detection.This article applies the SMOTE algorithm(synthetic minority oversampling technology)to the model data level,the Adaboost.M2 algorithm(improved Algorithm of Boosting in integrated learning)to the algorithm level,and the t-SNE-LSTM model as the weak classification in integrated learning.Form an intrusion detection model that enhances the ability to process unbalanced data.Experiments conducted on the NSL-KDD data set show that the t-SNE-LSTM model effectively classifies intrusion detection data.Compared with other models and methods,the comprehensive evaluation index F value of machine learning algorithms has been improved,96.274% and 96.668% on Normal and DOS.It is shown that after processing a few samples with the SMOTEAdaboost.M2 method,the model F value has been greatly improved,especially in the detection of the minority samples U2 R and R2 L,compared with the t-SNE-LSTM intrusion detection model,the F value is increased by 37.104% and 69.368%.