摘要
针对朴素贝叶斯(NB)算法在现实情况中所存在的缺陷,提出一种改进后的朴素贝叶斯算法——树加权朴素贝叶斯(TW-NB)算法。该算法通过引入决策树归纳法(DTI)在属性之间条件独立的集合中选择出相对更为重要的子属性集,并利用权重参数弱化了NB算法的条件独立假设性,从而降低了分类数据的维度,提高了算法的分类准确率。结合实验结果证明,在使用有限的计算资源下,基于TW-NB算法的入侵检测技术对于不同的网络入侵类型皆能表现出较高的检测率(DR)和较低的误检率(FR)。
Aiming at the deficiency of naive Bayesian （NB） algorithm in the reality, this paper proposes an improved NB algorithm which is called tree-weighting naive Bayesian （TW-NB） algorithm. This algorithm, by introducing the decision tree induction （DTI）, selects a comparatively more important subset of attributes from the set of conditional independence assumption, and uses weighting parameter to weaken the conditional independence assumption of naive Bayesian and thus reduces the dimensionality of the classification data, as well as improves the classification accuracy of the algorithm. It is verified by combining the experimental results that the intrusion detection technology based on the TW-NB algorithm can achieve higher detection rates （DR） and lower false rates （FR） for different network intrusion types when the computational resources used are limited.
出处
《计算机应用与软件》
CSCD
2016年第2期294-298,共5页
Computer Applications and Software
基金
国家自然科学基金项目(51174263)
教育部博士点基金项目(20124116120004)
河南省教育厅科学技术研究重点项目(13A510325)
