作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2006, Vol. 32 ›› Issue (17): 86-88. doi: 10.3969/j.issn.1000-3428.2006.17.030

• 专题论文 • 上一篇    下一篇

一种基于不完整数据的朴素贝叶斯分类器

陈景年1,2;黄厚宽1;田凤占1;乔珠峰1   

  1. (1. 北京交通大学计算机与信息技术学院,北京 100044;2. 山东财政学院信息与计算数学教研室,济南250014)
  • 收稿日期:1900-01-01 修回日期:1900-01-01 出版日期:2006-09-05 发布日期:2006-09-05

Naive Bayes Classifiers Learned from Incomplete Data

CHEN Jingnian1,2;HUANG Houkuan1; TIAN Fengzhan1; QIAO Zhufeng1   

  1. (1. School of Computer and Information Technology, Beijing Jiaotong University, Beijing 100044; 2. Dept. of Mathematics, Shandong University of Finance, Jinan 250014)
  • Received:1900-01-01 Revised:1900-01-01 Online:2006-09-05 Published:2006-09-05

摘要: 贝叶斯网络因其对属性间因果关系的表达能力而成为处理不完整数据的强有力的工具。然而绝大多数的贝叶斯分类器都是基于完整数据的,并且在现实世界中数据往往是不完整的,因此利用不完整数据构建有效的贝叶斯分类器是一个重要而又具有挑战性的问题。 通过分析著名的基于不完整数据的RBC分类器的不足,在BC (Bound and Collapse)方法和EM算法的基础上给出了一种基于不完整数据的分类器构建方法。实验结果表明了该算法的有效性。

关键词: 不完整数据, 贝叶斯分类器, EM算法, BC方法

Abstract: Bayes networks, as directed acyclic graphs of causal structure of attributes, have become the efficient method for processing incomplete data. However, most Bayesian network classifiers are learned from complete data and the world is rarely fully observable and data is often incomplete. So constructing Bayesian network classifiers from incomplete data is an important and challenging problem. An efficient method for constructing Bayesian network classifiers from incomplete data based on BC method and EM algorithm is presented. Experimental results show its validity.

Key words: Incomplete data, Bayes classifiers, EM algorithm, BC method

中图分类号: