The Research Of Bayesian Network On Intrusion Detection Based On Incomplete Data

With the rapid development of Internet technology,computer brought great changes to human and also play a pivotal role in the economic,cultural and other fields.But because of the operating system,web applications,hardware devices exist some security vulnerabilities that Often used by criminals to the user's property and privacy and security has brought a great threat.Computer security is becoming more serious,then Intrusion detection has become the focus of attention.The paper applies the Bayesian network,investigates for instrusion detection system.Bayesian network is a powerful tool for probabilistic reasoning.It not only reduce requirement of Na?ve Bayesian for conditional independence of attributes,but concisely reveal the dependency between attributes.Firstly,this paper elaborates current research status of intrusion detection and analysis the existing problem of intrusion detection;then it discusses the classification of intrusion detection systems and rough set.Finally,it deeply discuss the design and implementation of intrusion detection system based on Bayesian network and analysis its problem.For the incomplete data,the paper propose the R-BN algorithm.It applies the incomplete data analysis method of rough set to data completing.The principle of data completing is that the classification rules that complete data set make have a high degree of support and the classification rules are as concentrated as possible.The experiment of comparison analysis of classification efficiency between R-BN algorithm and conventional data completing algorithm.The traditional Bayesian network model can's make a real time response when the environment changes.It may affect classification efficiency.The paper proposes a R-BN algorithm base on dynamic structure change.It import sliding window and add the classified samples to the end of window.By sliding window the training set is constantly updated.When the environment changes,the scoring function determine whether to update the Bayesian Network parameters to improve classification efficiency.