Data-Centric AutomatedData Mining以数据为中心地自动化数据挖掘.pdf

《Data-Centric AutomatedData Mining以数据为中心地自动化数据挖掘.pdf》由会员上传分享，免费在线阅读，更多相关内容在教育资源-天天文库。

1、Data-CentricAutomatedDataMiningMarcosM.CamposPeterJ.StengardBorianaL.MilenovaDataMiningTechnologiesOverviewóDataminingcomplexityóProposeddesignsolutionóApplicationI:OraclePredictiveAnalyticsóApplicationII:SpreadsheetAdd-InforPAOverviewóDataminingcomplexityóProposeddesignsolutio

2、nóApplicationI:OraclePredictiveAnalyticsóApplicationII:SpreadsheetAdd-InforPADataMiningComplexityóKnowledgeofdataminingtechniques–WhichalgorithmdoIuse?óAlgorithmspecificdatapreparation–HowshouldIpreparemydata?óModelparametertuning–WhatkernelfunctionshouldIuse?óDeployment–I’vede

3、ployedamodel,nowwhatdatacanIscorewithit?IndustryFuture“Predictiveanalyticsbuildsonthedataminingmultistepprocessandstatisticalmodelingtechniquestoaddalayerofautomationandself-directedbuilt-inintelligence.Businessusers(andnotjustPh.D.statisticians)cannowanalyzelargeamountsofcusto

4、mer,supplier,employeeandproductdataforpatternsandtrends.”-KentBauerDMReviewMagazine,Dec.2005DesignApproachóGoal:“Goodresultswithminimumeffort”óData-centricfocus–familiartodatabaseandbusinessintelligencecommunitiesóProcessautomation–ease-of-usefornon-expertusersData-centricDesig

5、nóEliminatesconceptsofmodelsorcomplexmethodologiesóRequiresonlyknowledgeofthedatasourceóSupportingobjectsareeitherremovedorlinkedtothedatasourceóUsersseeonlypredictiveordescriptiveresultsóGoal-orientedtasksGoal-orientedTasksóExplain-attributeimportanceóPredict-classificationorr

6、egressionóGroup-clustering/segmentationóDetect-anomaly/outlierdetectionóMap-projectdatatolowerdimensionalityóProfile-supervisedsegmentationProcessAutomationóStatisticscomputationóSamplingóAttributetypeidentificationóAttributeselectionóAlgorithmselectionóDatatransformationóModel

7、creationandselectionóOutputgenerationProcessAutomation(cont.)óStatisticscomputation–numberofrecords,numberofattributes,attributerangesandcardinality–usedtomakedecisionsabouttargetandattributetypeandguidedatatransformationsóSampling(random&stratified)–improvetrainingtimesforlarg

8、edatasets–ensuresufficientraretargetvalue/rangereprese