Analysis And Prediction Of Intrusion Attacks Based On Deep Learning

With the rapid development and wide application of information technology,the network has penetrated into various fields of society.However,due to the openness,interconnectivity and diversity of the network,the network is vulnerable to attacks by malware,and network attacks occur frequently.Although security defense technology continues to improve,the traditional intrusion detection technology has little effect with the mutable attack methods.The rapid development of artificial intelligence in recent years has prompted the continuous updating of intrusion detection technology.This paper combines deep learning methods to detect and identify intrusion attacks to adapt to mutable attack behaviors and improve the recognition rate of intrusion detection.The main research contents of this article are as follows:(1)Analyze and process the NSL-KDD data for the model to calculate.The NSL-KDD data used in this paper is the simulation data of the laboratory.By analyzing the value range,correlation and probability distribution of each attribute,some problems were found.One is that the number of samples of each attack type is seriously unbalanced.This article alleviates the imbalance of the sample size by randomly oversampling the data.The second is that there is a certain difference in the probability distribution of the original train set and the test set.This paper discretizes some attributes of the train set and the test set and maps them to the same interval to reduce the difference in distribution.The experimental results show that the experimental results of the processed data are better.(2)Adapt the NSL-KDD data set by simplifying the residual neural network.As the number of neural network layers increases,the problem of gradient disappearance is easy to occur,and the residual block can effectively avoid this problem.But for the NSL-KDD data set,its complex structure will cause over-fitting problems.This paper reduces the complexity of the model and the dimension of the data by deleting some weight layers of the original residual block and adding a pooling layer.At the same time,the PRe LU function is used to replace the Re LU function to prevent neurons from "death".Finally,a neural network model is constructed by cascading the simplified residual blocks.The experimental results show that the recognition rate of the constructed model on NSL-KDD is higher than that of the original residual block.(3)Improve the recognition rate of intrusion attacks by ensemble neural networks.Ensemble learning can improve the imbalance of sample size and improve the overall detection rate of the model.This paper combines neural networks and ensemble learning methods.By using multilayer perceptron,recurrent neural networks,and convolutional neural networks as sub-models and multilayer perceptron as meta-learners,an ensemble neural network model is built using the Stacking integration method.The experimental results show that the detection rate of the ensemble neural network model on NSL-KDD is improved and the recognition rate of some categories is also improved.(4)Perform cluster analysis on intrusion attack data to discover unknown types of attacks.Cluster analysis can find similarities and differences between samples.Samples with high similarity are likely to belong to the same category;samples with large differences may be new types of attacks.In this paper,through the incomplete autoencoder model to reduce the dimensionality of the data and extract important feature information.Then the Mean Shift algorithm is applied to perform cluster analysis on the dimensionality-reduced data.The experimental results show that cluster analysis can help researchers find unknown attack types,and the cluster algorithm still needs to be improved.