Research On Key Technology Of Intrusion Detection Based On Naive Bayesian

With the rapid development and application of network technology,and frequent exposure of security issues worldwide,such as monitoring,and malicious attack,more and more attention is paid to the intrusion detection system,for it is the key of weather computer technology could develop sustainably.Intrusion detection system(IDS)is a barrier to protect the network.The core of IDS is intrusion detection algorithm.Naive Bayes algorithm is often used in intrusion detection because of its strong classification ability.However,in practice,Naive Bayesian still has problems and should be improved.In this paper,research is established on attributes' independence,continuous attributes' modeling and sample quality.The improvements are as follow:(1)Restricted Boltzmann machine-Na?ve Bayes classifier model is constructed in this paper to enhance independence between data's attributes.Na?ve Bayes classifier is based on Bayesian,and it assumes that data's attributes are independent of each other.However,it is difficult to set up in practice.Restricted Boltzmann machine could reduce correlation between attributes during dimension reduction,and this could improve the detection accuracy of Na?ve Bayes classifier theoretically.(2)The improved Na?ve Bayes classifier based on Gaussian distribution is proposed for continuous attributes' modeling.Na?ve Bayes classifier is skilled in dealing with discrete data,and is weak in dealing with continuous data.A clustering method is defined to solve continuous data's modeling,which could explore the cluster centers of continuous data.Then,Gauss model is used to model the continuous data,to improve the accuracy of the conditional probability in training phase.(3)Na?ve Bayes classifier based on linked attributes is proposed to Improve the classifier's detection ability on samples with poor quality.First,explore the attributes which could lead classification errors by analyzing the error items of the initial classification.Then set linked attributes with weights and expand the samples.At last,adjust the weights according to probability variance given in this paper to improve the adaptability of the classifier in complex situation.