摘要
树扩展朴素贝叶斯分类器(TANC)是应用较广的一种贝叶斯分类器。TANC的分类性能优于朴素贝叶斯分类器(NBC)。现有的TANC结构学习算法是基于相关性分析的,采用互信息测度。贝叶斯信息测度(BIC)在基于打分和搜索的贝叶斯网络结构学习中取得了成功,文中用BIC测度来衡量属性结点之间的相关性,提出了一种新的TANC-BIC结构学习算法。在MBNC实验平台上编程实现了TANC-BIC算法,用分类准确率衡量算法的性能。实验结果表明,TANC-BIC算法是有效的。
TANC, which is one type of Bayesian classifier,is applied widely. TANC is superior to NBC. Existing TANC structure-learning algorithm is based on relativity analysis using mutual information criterion. BIC has success with structure-learning of bayesian networks based search & scoring. This paper suggests a new TANC-BIC structure-learning algorithm which using BIC computes nodes relationship. TANC-BIC has programmed under MBNC experiment platform. Using classification accuracy scales classification performance. Experiment results show that TANC-BIC is effective.
出处
《微机发展》
2004年第11期10-12,共3页
Microcomputer Development
基金
清华大学智能技术与系统国家重点实验室开放课题资助(99002)