基于互信息可信度的贝叶斯网络入侵检测研究

传统贝叶斯入侵检测算法没有考虑不同属性和属性权值对入侵检测结果的影响,因此分类准确率不够高.针对传统贝叶斯入侵检测算法存在的不足,提出基于互信息可信度的贝叶斯网络入侵检测算法.在综合考虑网络入侵检测数据特点和传统贝叶斯分类算法优点的基础上,用互信息相对可信度进行特征选择,删除一些冗余属性,把互信息相对可信度作为权值引进贝叶斯分类算法中,得到优化的贝叶斯网络入侵检测算法(MI-NB).实验结果表明,MI-NB算法能大大降低分类数据的维数,比传统贝叶斯入侵检测算法及改进算法有更高的分类准确率.

Bayesian network intrusion detection method based on credibility of mutual information

Traditional Bayesian intrusion detection algorithm does not consider the influence caused by different properties and weights of the properties, so the classification accuracy rate is not high enough. Aiming at the shortage of traditional Bayesian intrusion detection algorithm, a Bayesian network intrusion detection method based on credibility of mutual information is proposed. After considering the characteristics of network intrusion detection data and the merits of traditional Bayesian classification, credibility of mutual information is used to select feature, and some redundant properties are deleted. The credibility as weights is introduced Bayesian classifier in order to get optimized Bayesian network intrusion detection algorithm (MI-NB). Experiments show that MI-NB algorithm can greatly reduce the dimension of classification data and has higher classification accuracy rate than the traditional intrusion detection algorithm and the improved algorithm.