Research On Intrusion Detection Technology Based On NBSR Model

Nowadays,under the circumstances of the high-speed and interconnected big data network,diversified network attack methods has made more frequent network security incidents,causing great losses to citizens,society and even the country.As traditional network intrusion detection approaches have been far from enough to meet the demand of safeguarding the security of big data network environments,the combination of machine learning and intrusion detection has become a new research direction under the increasingly severe security situation in cyberspace.At present,many intrusion detection classifiers based on machine learning technology usually neglect the comprehensiveness of data set in training,which leads to a high false negative rate of classifiers.Therefore,starting with the research on intrusion detection data sets,this dissertation focuses on improving the overall level of machine cognition,and proposes an intrusion detection technology based on NBSR model.Based on fully studies the characteristics of the existing intrusion detection data sets,the author selects the representative UNSW-NB15 as the training and testing data set around which all the research work has been carried out.The main work and innovations of this dissertation are as follows:(1)The Relief F feature selection algorithm was introduced to solve the problem of large feature dimension of UNSW-NB15 dataset in this dissertation.The Pearson correlation coefficient were introduced to make up for the lack of correlation analysis of Relief F feature selection algorithm.The Relieff-P algorithm was proposed to remove noise and boundary data,and weakened the correlation between features.(2)In view of the shortcomings of the traditional Naive Bayesian classifier,Softmax regression classifier and its cascade are proposed to form NASS classifier,and the performance of Naive Bayesian classifier is improved by dispersing sparse samples to weaken the influence of migration.On the base of the above analysis,an NBSR model was established,and the NASS classifier was trained by using the UNSW-NB15 data set through the "Supervised Learning(SL)" method.The experimental results showed that the feature dimension can be reduced to 8-D at the lowest level,and the false negative rate of NBSR model is only 15.62% which is lower than that of other intrusion-detection models.