一种基于不完整数据的朴素贝叶斯分类器  

Naive Bayes Classifiers Learned from Incomplete Data

作　　者：陈景年[1] 黄厚宽[1] 田凤占[1] 乔珠峰[1] 

机构地区：[1]北京交通大学计算机与信息技术学院

出　　处：《计算机工程》2006年第17期86-88,共3页Computer Engineering

基　　金：国家自然科学基金资助(60503017)

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(BoundandCollapse)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。Bayes networks, as directed acyclic graphs of causal structure of attribute s, 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.

关 键 词：不完整数据 贝叶斯分类器 EM算法 BC办法 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]