利用粒分辩关系构建决策树的算法探讨.pdfVIP

ComputerEngineering andApplications计算机工程与应用 @数据库、信号与信息处理@ 利用粒分辩关系构建决策树的算法探讨 刘 军 LIUJun 南京工业大学电子与信息工程学院，南京210009 ofElectronicsandInformation University 210009，China College Engineering，NanjingofTechnology,Nanjing LIU on of decisiontree constructing granulationrelation．Computer Jun．Studyalgorithm using and Appfications，2012，48(26)：109-112． atthe ofinaccurate adecisiontreebasedontheinformation Abstract：Aimingproblem constructing gainalgorithm to the attributesandall ofthe set coreattributesasthe attri- determine preferred using preferred algorithmrough toitsnaturevaluetheattributeisdividedintoseveralbasic butes，ingranulationtheory,according granulation． conditional volumeof theresolutionof tode- granulationresolution)and granulation Through length(the relativing of selection issolvedUS— cision class attributeisdivided．Multi—attribute resolution)，the problem by granulation(the the and resultsshowthatthe achieves andformula simpleformula．Theoryexample algorithmprecision ing analysis is effectiveandthe and islow． simple temporalspatialcomplexity set；decision relation Keywords：rough tree；granulatio 摘要：针对信息增益算法以信息量的大小确定首选属性和基于粗集算法以核属性为首选属性构建决策树的 不准确问题，以粒理论为基础，将属性按其自然取值划分为若干基本粒，以条件粒的长度(粒分辩量)和该粒对 决策粒分辩关系(分辨类别)为依据确定划分属性，采用简洁的算式解决了多属性的择优难题。理论和实例分 析的结果表明，该算法具有建树精准简洁有效以及时空复杂度低的特点。