连续属性朴素贝叶斯分类器的依赖扩展研究

The dependency extension of naive Bayes classifiers with continuous attributes

针对朴素贝叶斯分类器不能有效利用属性之间依赖信息的问题,在将连续属性条件互信息计算、条件密度计算与通过建立类约束属性最大权重跨度树的父结点选择相结合的基础上,提出了连续属性朴素贝叶斯分类器选择性树结构依赖扩展方法.通过对比实验和分析,证实了扩展后分类器的分类准确率得到明显的改进.

On account of naive Bayesian classifiers can＇t make good use of the dependence information between attribute variables. We extend naive Bayesian classifiers with continuous attributes using treelike graphical models. The method based on computation of mutual information of the continuous attributes and conditional density, combining the construction of parent node selection of the class- constrained attribute maximum weighted spanning tree. Comparative experiments and analysis are done. Experimental results show that classification accuracy of the extended naive Bayesian classifier with continuous attributes has improved obviously.