Journal of Beijing University of Posts and Telecommunications

  • EI核心期刊

JOURNAL OF BEIJING UNIVERSITY OF POSTS AND TELECOM ›› 2012, Vol. 35 ›› Issue (1): 72-76.doi: 10.13190/jbupt.201201.72.wangzhf

• Papers • Previous Articles     Next Articles

Equivalent Classed of TAN Classifier Structure and Their Application on Learning Algorithm

WANG  Zhong-Feng1,2, WANG  Zhi-Hai2   

  1. 1.School of Computer and Information Technology, Bijing Jiaotong University, Beijing 100044,China 2.Safety and Emergency Management Laboratory, Beijing Municipal Institute of Labour Protection, Beijing 100054,China
  • Online:2012-02-28 Published:2012-02-28

Abstract:

As the structure of a tree augmented Naive Bayes(TAN)classifier is a directed acyclic graph, most of learning algorithms try to improve classifier's performance by selecting the directions of arcs. With scoring criterion of log conditional likelihood, however, it is proven that TAN classifiers with same links are equivalent after analyzing the searching space and equivalent classed of TAN classifier structures. Therefore, a learning algorithm without taking arc reversal into account is proposed. Experienments show that the proposed altorithm is effective.

Key words: data mining, classifier, Bayesian network, equvalent classes, structure learning

CLC Number: