Research On Intrusion Detection Based On Improved Unbalanced Strategy And Deep Learning

The rapid development and widespread popularization of the Internet has facilitated human life,and the rich information contained therein has brought many network security issues.In today's complex network environment,various viruses,vulnerabilities and attacks emerge in an endless stream,threatening social public security and causing serious economic losses.Therefore,protecting network security has become an urgent research topic.Intrusion detection can detect attacks proactively,so it plays a key role in network security.With the continuous evolution of network intrusion attacks,the drawbacks of traditional intrusion detection methods are increasingly apparent,resulting in poor detection performance,low efficiency,and difficulty in identifying rare categories.The emergence of deep learning can effectively solve this problem,and deep learning has been successfully applied in many fields before.The multi-layer network structure of deep learning and excellent feature learning ability contribute to the extraction of complex data features in intrusion detection.In this paper,an in-depth analysis of intrusion detection methods and deep learning models is carried out to study intrusion detection based on deep learning.The main work of this paper is as follows:In view of the fact that traditional intrusion detection technology is difficult to extract high-dimensional data,resulting in low recognition rate,this paper proposes an intrusion detection model DBN-PNN based on Deep Belief Networks(DBN)and Probabilistic Neural Network(PNN).The model utilizes the powerful feature learning ability of DBN combined with the advantages of fast convergence and fault tolerance of PNN to realize the detection of intrusion behavior.DBN-PNN first uses DBN to reduce the dimension of a large number of original high-dimensional data,completes the feature extraction,and then uses PNN to classify the low-dimensional data output by DBN.The experimental results show that the model has a good recognition effect.In view of the uneven distribution of network intrusion data,the proportion of each category is seriously unbalanced,most intrusion detection methods only focus on the detection rate of the whole sample,ignoring the problem of a few categories.This paper adds optimization processing to unbalanced data based on DBN-PNN model.A deep learning intrusion detection model based on optimized unbalanced data is proposed.There are four modules in this model.This paper studies the data preprocessing module and intrusion detection module.In the data processing module,the Synthetic Minority Over-sampling Technique algorithm(SMOTE)is used to synthesize a few categories of samples,and the number of categories is increased.The Neighborhood Cleaning Rule algorithm(NCL)is used to under-sampling most categories of samples to reduce the number and eventually form balanced data.DBN-PNN proposed in this paper is applied to the intrusion detection module.Finally,the experimental results verify that the model can effectively improve the overall detection effect and enhance the identification of a few types of attacks.In summary,the model proposed in this paper not only effectively improves the detection performance of intrusion detection,but also reduces the detection time.At the same time,for the improvement of non-equilibrium strategy,the model of this paper can significantly improve the accuracy of a few types of attacks.