The Research On Efficient And Practical Method For Intrusion Detection Based On IG-PCA And Compressed Sensing

In recent years,with the rapid expansion of the network scale,there existing the massive network traffic be analyzed by network intrusion detection equipment,However,these network data sets have the characteristics of high dimension and large redundancy.The existing methods such as feature selection and clustering have achieved some results in the simplified calculation,but there are still problems to be solved like slow convergence speed and low detection accuracy.Feature reduction and sample reduction are commonly used methods in data preprocessing,studying an effective data preprocessing algorithm to achieve the purpose of real-time and efficient intrusion detection is significant and has a wide range of application prospects.In this thesis,the research status of network intrusion detection based on machine learning is first introduced,and then describes detailedly the basic framework of intrusion detection and the method of feature reduction and sample reduction in data preprocessing.Finally,we combine the IG-PCA algorithm used in feature dimension reduction process with CS algorithm conducted in the process of sample reduction,a feedback intrusion detection model based on IG-PCA and compressive sensing is proposed.The main works of this paper are as follow:(1)On the basis of the traditional feature selection and feature extraction method,this paper proposes a feature dimension reduction method based on Information Gain and PCA.First of all,this method uses the Gain Information algorithm to sort the features of the samples.Then,the PCA algorithm is conducted to determine the feature dimension that needs to be retained,thus,we can select the optimal feature subset.(2)The compressed sensing algorithm in the field of image processing is introduced into intrusion detection,through compressed sampling of the original repeated sample data set,the constructed small sample set can greatly simplify the calculation so as to shorten the detection time.And by calculating the error of the correct classification rate before and after the sample reduction to determine whether this process is an effective compression.This experiment shows that the sample reduction method is effective for intrusion detection.(3)A feedback intrusion detection model based on IG-PCA and compressed sensing is proposed,which is based on the feature dimension reduction and sample reduction method.Through observing the accuracy of the final results and choosing different observation matrix rows for compression to control the times of sampling,we can obtain the optimal sampling rate,so as to achieve the purpose of real-time,high precision detection.The method proposed in this paper can greatly shorten the detection time under the condition that the correct classification rate is relative high,which can be widely used in a variety of intrusion detection systems.