散度加权的平均一阶依赖估计分类算法研究  

Averaged one-dependence estimators classification algorithm based on divergence weighting

作　　者：陈圣磊 高兴宇 卓超 朱昌舰 Chen Shenglei;Gao Xingyu;Zhuo Chao;Zhu Changjian(School of Economics,Nanjing Audit University,Nanjing 211815,China;School of Computer Science,Nanjing Audit University,Nanjing 211815,China)

机构地区：[1]南京审计大学经济学院,江苏南京211815 [2]南京审计大学计算机学院,江苏南京211815

出　　处：《南京理工大学学报》2024年第4期479-488,共10页Journal of Nanjing University of Science and Technology

基　　金：国家自然科学基金(62276136);江苏省研究生科研与实践创新计划项目(SJCX23_1105)。

摘　　要：平均一阶依赖估计(AODE)是对朴素贝叶斯分类算法的重要扩展,然而,AODE平等地对待各个属性,这限制了其分类性能的提升。为了准确刻画各个属性对于分类的作用,进一步提升AODE的分类性能,该文提出一种基于散度加权的AODE分类算法。该方法引入了Kullback-Leibler散度和Jessen-Shannon散度2种散度指标,基于类别的先验分布和给定属性取值的后验分布之间的散度,构建AODE分类框架中超级父属性一阶依赖估计器的权值,从而得到超级父属性一阶依赖估计器的更优组合方式。在36个加州大学机器学习数据集上的实验表明,基于散度的AODE属性加权算法显著优于原始的AODE算法。因此,散度加权能够有效提升AODE的分类性能。The averaged one-dependence estimators(AODE)algorithm is an important extension of the naive Bayesian classification algorithm.However,AODE treats all attributes equally,which limits its ability of improving classification performance.In order to accurately characterize the role of each attribute in classification and further improve the classification performance of AODE,this paper proposes a divergence weighted AODE classification algorithm.The method introduces two divergence metrics,Kullback-Leibler divergence and Jessen-Shannon divergence,and uses them to determine the weights of the super parent one-dependence estimators in the AODE classification framework.These weights are based on the divergence between the prior distribution of the class variable and the posterior distribution of the given attribute values.The results are in a more optimal way of combining the super parent one-dependence estimators.Experiments on 36 data sets from the University of California machine learning repository show that the divergence weighted AODE algorithm significantly outperforms the original AODE algorithm.Consequently,the use of divergence weighting can effectively improve the classification performance of AODE.

关 键 词：平均一阶依赖估计 Kullback-Leibler散度 Jessen-Shannon散度 加权 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]