数据归约的统计方法研究及应用

《数据归约的统计方法研究及应用》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、厦门大学博士学位论文数据归约的统计方法研究及应用姓名：刘云霞申请学位级别：博士专业：统计学指导教师：曾五一20070301实例验证了该方法的有效性。第六章对全文的主要工作进行了总结，并指出了有待进一步改进和完善的地方。本文的创新之处主要有以下四个方面：(1)提出了分别基于可辨识矩阵和基于似然比假设检验的两种连续属性离散化方法。(2)提出了单向有序列联资料属性排序的方法——改进秩和法。(3)提出了基于因子分析的无监督属性重要性的排序方法。(4)提出了逐步向前的无监督属性选择方法。关键词：数据归约；数据挖掘；统计学Abst

2、ractDatareductionisthekeystepofDataMininganditisimportanttostudythemethodsofdatareduction．Majorityofexistingmethodspaymoreattentiontosupervisedlearningcurrently．HoweverthestudyoftheunsuperviseddatareductionWaSh’tabundantrelatively．Thereforethisdissertationfocuse

3、sonthestudytothestatisticalmethodsandapplicationoftheunsuperviseddatareduction．InChapterone，thebackgroundsandsignificanceoftheselectedtopicwereillustratedfirstly．Afterwards，onthebasesofsummarizingrelevantbackgroundsandstudymethodsofthedatareductionfrombothhomean

4、dabroad，wepointedoutthecontentsandtheinnovativeplacesofthispaper．InChaptertwo，itWasdiscussedthemissingvalueimputationandtheoutliersdetectionwhicharethebaseworkofdatareduction．Inthischapter,wesummarizedsomemethodswhichcanbeappliedinDataMingonthebasisoftheanalysis

5、tothosestatisticalmethods．Inaddition，weanalyzedconsumers’consumptivebehaviorbythemethodsoftheoutliersdetectionappliedinthedatabaseofthesomeconsumptivemobiletelecommunication．Datareductionincludestuplesreductionandattributesreduction．InChapterthree，wediscussedthe

6、discrezationofcontinuousattributesandtheconcepthierarchywhicharetwomainmethodsoftuplesruduction．Onthebasesofthesummaryofthecurrentmethodsofthediscrezationandattributeorientedinduction，weputforwardtwomethodswhichwerethediscretizationofcontinuousattributesbasedond

7、iscernibilitymatrixandthediseretizationofcontinuousattributesbased011likelihoodratiohypothesistesting．ThesimulationtothesemethodsintheIrisdatabasevalidatedtheirvalidation．Themethodsofattributesreductionincludetheimportanceorder,theextractionandtheselectionofattr

8、ibutes．InChapterfour,wediscussedtheimportanceorderofattributes．ThesupervisedimportanceorderofattributesisfamiliarinDataMining．Wefirstly,madeanintroductiontoit．Andthen