一种多值属性多类标数据决策树算法  

A Multi-Valued Attribute and Multi-Labeled Data Decision Tree Algorithm

作　　者：李宏[1] 陈松乔[1] 赵蕊[1] 郭跃健[1] 

机构地区：[1]中南大学信息科学与工程学院,长沙410083

出　　处：《模式识别与人工智能》2007年第6期815-820,共6页Pattern Recognition and Artificial Intelligence

基　　金：国家自然科学基金(No.60433020)

摘　　要：目前处理多值属性多类标数据的算法有多值多类标分类器(MMC)和多值多类标决策树(MMDT).本文在研究前面两种算法的基础上提出新的相似度计算公式 sim_3,并通过改进 MMDT 基于一致性的评定方法,提出一种处理多值属性多类标数据的算法 SCC_SP,综合考虑两个多类标集合的相似性和一致性,更有利于选择最佳分裂属性.通过对比实验证明,在相同的预测机制下,SCC_SP 的预测准确度比 MMDT 高,能更好地处理多值属性多类标数据.Multi-valued and multi-labeled classifier （MMC） and multi-valued and multi-labeled decision tree （MMDT） are two existing decision tree algorithms for dealing with multi-valued and multi- labeled data. Based on the two algorithms, formula sim3 is put forward to calculate the similarity between two label sets. By amending the measuring formula of same-based similarity of label-sets in MMDT, a new decision tree algorithm, similarity of same and consistent in constructing same in predicting （SCC_SP） is proposed with comprehensive consideration of both similarity and appropriateness of the label set. Results of contrast experiments with the same prediction mechanism show that SCC_SP has higher accuracy rate than MMDT.

关 键 词：分类 决策树 多值属性 多类标数据 相似度 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]