Research On Classification Algorithm Of Intrusion Detection Based On Machine Learning

As an important technology to ensure information security,intrusion detection has received great interest from researchers.Nowadays,the Internet has become a necessity in people's daily life.However,with the development of the Internet,cyber-attack has become increasingly difficult to detect and it has become more harmful than before.In recent years,machine learning and deep learning have made great achievements in data processing and data prediction.Moreover,lots of researchers who are interested in intrusion detection have been inspired by these achievements.Machine learning and deep learning can select normal data and abnormal data among the massive data,and they provide new methods for intrusion detection.In the process of processing and classifying data,the traditional intrusion detection technique often uses a single classifier.Hence,it is difficult to accurately classify all types of data and detect new types of cyber-attacks.Some feature selection methods,which are widely applied in the process of processing and classifying data,often ignore the correlation among data.Therefore,the accuracy of further classification is affected.In order to solve these problems,the paper does as follows:(1)As for data with multiple types,this paper proposes an intrusion detection technique based on multiple classifiers with multiple steps of classification.This technique establishes a multi-class model,improves the feature selection method,and focuses on the correlation between features.In addition,a classifier with the best classification effect for a certain type is used to complete the data classification in the process of analyzing multiple types of data.The result of experiments on KDD CUP99 dataset shows that this technique has significant effect on multi-class classification.(2)In order to improve the detection accuracy,an intrusion detection method based on CDBN(convolution depth belief network)is proposed.First,the technique converts character-type data features into numerical features.Second,it expands the dimension of feature extension and transform its one-dimensional vector feature into two-dimensional matrix feature.Third,it adds the batch normalization layer(BN)behind the convolutional layer of the convolutional neural network in order to accelerate the training speed of the network and prevent the model from over fitting.Last,the identification and classification of the data are finished by CDBN.The result of the simulation experiment,which is proceeded on KDD CUP99 dataset,indicates that this technique improves the classification accuracy of intrusion detection.This paper makes in-depth and detailed research on intrusion detection from the aspects of machine learning and deep learning,paves the way for the research based on classification algorithms in machine learning,and vindicates the importance and feasibility of the research through the experiment.