描述性规则挖掘若干关键技术研究

《描述性规则挖掘若干关键技术研究》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、I1040716分类号：至￡3U密级：公珏UDC．512：鳢学号：．Q22123东南大学博士学位论文描述性规则挖掘若干关键技术研究研究生姓名：瞳金壬导师姓名：处圭捏一一夔援申请学位级别王堂墟±学科专业名称j士篡扭座周撞苤论文提交日期2Q笾：12：2Q论文答辩日期学位授予单位苤宙盍堂学位授予日期窒QQ鱼生且目答辩委员会主席评阅人2006年2月22日AbstractDescriptivepatternminingis强importantissueindataⅡljning．mtaskofdataminingist

2、odiscoverpatternsinmassivedata,itc姐bedividedintotwocalegori鹳accordingtothefunctionofpatterns，thatispredictivepatternmininganddescriptivepatternmining．DescriptivepattemminingmakesdescriptiontothepotentialmlesjIlthedatas吐whichcontainssomeimportantre∞,archiSSU

3、eofdatamining,suchasassociationrulemining,sequentialpatternminingandsOon．ConsideringthepotentialrequirementsoftheconstructionofJiangsuintellectualpropertypublic5ea'vieec2nmrandaccordingthenseofpatentdataandexistedresearchindescriptivepaRemmining,theresearch

4、workisdoneinsequentialpenemmining．associ拍onrulesminingindistributedenvironmentandoutlierdetectionbaSedonaSsociationanalysis．TheprocessofsequentialpatternminingisrepeatedandmntuaJ，mefficientmntnalalgorithmbaSedonPrefixSpanisproposed，whichcallmakefullnseofthe

5、previousresultsandgeneratenewpatternsefficientlywhentheminimumsupportthresholdischanged；consideringtheproblemofincrementalupdatingofsequentialpattern,algorithmISPMP(IncrementalSequentialPatternsMiaingBasedProjectedDatabase)isproposed．whichupdatesthefrequent

6、itemsandpatternsfoundprevionslybyimplicitmerglnganddiscoveriesflewpatternsbyprojectiondatabase；furcllermore，toimprovetheefficiencyofsequentialpaaernmining，themethodofbinarycodingisappliedtogeneratecandidatesequentialpatternandcomputecorrespondingcounts，whic

7、honlyneedsomesimpleoperationas‘‘or'’，‘‘and’’and‘'xor'’．Consideringtheproblemofassociationrulemininginmassivedatasets，analgorithmoffastminingglobalmaximumfrequentitemsets(FMGMFDisproposed,whichcarloonveniantlygettheglobalfrequencyofanyitemsetfromthecorrespon

8、dingpathsofeverylocalFP-treebyusingfrequentpatterntree，andrequiresfarlesscommunicationoverheadbythesearchingstrategyofbottom—upandtop-down．Finally,consideringtheproblemthatmostofcurrentoutlierdetection