Research On Intrusion Detection Based On Deep Learning

With the development of information technology,the Internet has gradually involved all aspects of life.People's lives are increasingly dependent on the network,and the network brings us convenience and also makes us face network security issues.In recent years,network security incidents have exploded,and people are increasingly aware of the importance of network security.Any network intrusion can cause irreparable disasters.As a passive defense mechanism,intrusion detection technology plays a very important role in network security.Traditional intrusion detection is mostly based on rules matching and statistics.With the advent of the big data,traditional intrusion detection has not been able to get excellent performance,especially facing massive,complex and unbalanced intrusion data.In the current environment,how to improve the overall performance of intrusion detection is a major challenge in this field.After fully understanding the related knowledge of intrusion detection,unbalanced data set and deep learning,this paper proposes an intrusion detection model based on convolutional neural network,which provides a new solution for intrusion detection.The contributions and main research contents of this paper are as follows:1.Detailed analysis of intrusion detection data sets such as KDD CUP 1999 and NSL-KDD.It is found that there are drawbacks in the KDD CUP 1999 data set,and the NSL-KDD data set solves this drawback,but the label distribution imbalance exists in the NSL-KDD data set.Aiming at the imbalance of label distribution,this paper proposes a method,which solves the problems caused by data imbalance from two aspects of data and algorithm.Using oversampling techniques at the data level,using oversampling techniques can increase the number of samples in a few categories,make the distribution of sample labels reasonable;using the Focal Loss function at the algorithm level,which is a loss function based on cost-sensitive vectors,which solves the problem of imbalance in classification problems,different difficulty levels between different categories,etc.2.Deep learning has developed rapidly in recent years and has achieved good results.Convolutional neural network,as an algorithm of deep learning,has achieved great success in the fields of image processing and natural language processing.Therefore,this paper applies the convolutional neural network to the field of intrusion detection,and proposes the corresponding intrusion detection model combined with the actual situation.The model uses threshold convolution,small convolution kernel,Dropout,Softmax and other methods.Based on the characteristics of the data set and the actual classification of the model,the corresponding optimization strategy is proposed.3.Accuracy is often used as an evaluation indicator for the performance of the model in the balanced data set,but the accuracy cannot be considered oqnly in the unbalanced data set.The dataset used in this paper is unbalanced.Therefore,when measuring the quality of the model,the model will be comprehensively evaluated from the four planes of accuracy,precision,recall,and F1 score.After a detailed analysis of the current research status,it is found that there are two evaluation model methods based on training set and test set.This paper will also conduct experiments from these two evaluation methods and compare with the existing literature.The five-category experiments on the NSL-KDD dataset show that the proposed intrusion detection model based on convolutional neural network improves the accuracy and different types of detection.As result,the model proposed in this paper has achieved good results.Compared with other algorithms,this paper has a higher detection rate on two types of attacks such as U2L and R2L.