孤立點挖掘及其內(nèi)涵知識發(fā)現(xiàn)的研究與應(yīng)用.pdf

1、Y7847a1廣西師范大學(xué)碩士研究生學(xué)位論文孤立點挖掘及其內(nèi)涵知識發(fā)現(xiàn)的研究與應(yīng)用ResearchandApplicationofOutlierMiningandFindingIntentionalKnowledge院系：年級：學(xué)科專業(yè)：研究方向：導(dǎo)師：研究生：完成時間；數(shù)學(xué)與計算機科學(xué)學(xué)院2002級計算機軟件與理論數(shù)據(jù)庫林士敏教授陸聲鏈2005年4月廣西師范大學(xué)碩上研究生學(xué)位論文ResearchandApplicationofOutl

2、ierMiningandFindingIntentionalKnowledgeAuthor：LuShenglianTutor：ProfessorLinSiftminSubjeet：ComputerScienceGuangxiKeyLabonInformationIntelligence＆DataProcessingAbstractTheproblemofoutliermininghasbeenvariouslycalledoutlier

3、analysis，anomalydetection，exceptionmining，detectingrareevents，miningrareclasses，deviationdetection，etcOutliermaybe“dirtydata’，butitalsocanmeansmeaningfuleventcorrespondingtotherealityFromthepointofknowledgediscovery，rare

4、eventareoftenmoreinterestingandvaluablethanothersinmanydomains，wheretherareevents’importanceisquitehighcomparedtootherevents，makingtheirdetectionandanalysisextremelyimportant，Themainworkinthethesisare1isted∞follows：(1)Su

5、mmarizingtheproblemofoutlierminingfromtherealisticmeaningalgorithms，applicationranges，detectiontools，algodthm’sevaluation，etc(2)Toovercomethelimitationofrequiringthresholdinexistingdistancebasedalgorithmsthispaperpropose

6、sanewdefinitionofoutlier弛isdefinitionUSeoneobject’sdistancewith“Iothersobjectsinthedatasettojudgetheobjectifalloutlierornot，sotheproblemthatthealgorithmneedssettingthenearneighbourparameterPorkissolvedeffectivelyToimprov

7、etheefficiencyasamplingbasedapproximatedetectionalgorithmhasbeendevelopedExperimentshavebeencarriedoutwithrealdata，theresultsindicatesthatnotonlythenewlydefinitiongetthesameresultsasDBp，∞’s，butalsopointsouttheoutlier’sou

8、tlyingdegreeinthedataset，andsimplifiestherequirementfordetectingoutliers(3)Thispaperresearchestheproblemoflocaloutlierinmutildimensionalandcategoricalattributedatasets，andgiveanewdefinitionofoutlierbaseonthetheoryofdiffe

9、renceinfrequentlyoftheattributevalue，andproposetheanomalyexistingcriterionforestimatingthesignificanceofthedetectedoutlierTheexperimentalresuaindicatesthatthecriterioncangetridofalargeamountofobjectswhichdegreeofdeviatio

10、nisnotremarkablereallyWeeveluatethevalidityofouralgorithmfromfouraspects：interestingoftheresult，comparisonwithsimilaralgorithms，contributionstotheimprovingoftheaccuracyrateofclassificationandabilityofdetectionrareclass(4

11、)AnexperimentalplatformnamedSOD(SmartOutlierDetection)hasbeenconstructed，whichintegratesthealgorithmsproposedorimprovedinthispaperandprovidesatoolforiheanalysisofoutlierdetectionItcan01：taiudatafromseveraloutaidedata$otl

12、rcesbythedataaccessinterface，thisstrengthensitspracticability，SODhasbeenintegratedwithateachingmanagementsystem(5)Combiningthecharacteristicoftheteachingmanagementsystem，thispagediscussesthenecessityforusingoutlierdetect