final presA Density basedalgorithm for Discovering Clusters in Large Spatial Databases with Noise

《final presA Density basedalgorithm for Discovering Clusters in Large Spatial Databases with Noise》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、ADensitybasedalgorithmforDiscoveringClustersinLargeSpatialDatabaseswithNoisebySreeLakshmiPeriOutlines•Problemdescription.•Contributions.•Keyconceptsbehindthepaper.•Densitybasednotionofclusters.•IntroductiontoDBSCAN.•Outlineofthealgorithm.•PerformanceEvaluation.•Results.•Conclusion.•FutureWork

2、.•IfIrewritethepaper.•Reference.ProblemDescriptionApplicationsoflargespatialdatabasesrisessomerequirementsforclusteringalgorithmssuchas:DomainknowledgetodeterminetheinputparametersDiscoveryofclusterswitharbitraryshapeandGoodefficiencyonlargedatabasesetsExistingalgorithmscan’tgiveasolutionforc

3、ombinationfortheserequirements.Henceanewapproachisrequired.Contributions•Effectivelydiscoversclustersofarbitraryshape.•Canbeusedforlargespatialdatabases.•Discoversnoiseandhandlesiteffectively.KeyConceptsbehindthepaper:Clustering:Clusteringistheclassificationofobjectsintodifferentgroups,ormore

4、precisely,thepartitioningofadatasetintosubsets(clusters),sothatthedataineachsubset(ideally)sharesomecommontrait.Itisatechniqueusedindatamining.Clusteringalgorithms:Therearetwokindsofclusteringalgorithms.•Hierarchicalclusteringproceedssuccessivelybyeithermergingsmallerclustersintolargerones,or

5、bysplittinglargerclusters.•Partitionalclusteringattemptstodirectlydecomposethedatasetintoasetofdisjointclusters.KeyConcepts(Contd.)Densitybasedapproach:Clustersareregardedasregionsinthedataspaceinwhichtheobjectsaredense,andwhichareseparatedbyregionsoflowobjectdensity(noise).Theseregionsmayhav

6、eanarbitraryshapeandthepointsinsidearegionmaybearbitrarilydistributed.DensitybasednotionofclustersForanycluster,wehave:•Acentralpoint(p)•Adistancemetricfromthepoint(Eps)•Minimumnumberofpointswithinthespecifieddistancemetric(MinPts)Foragivenpoint‘p’,thepointscontainedwithinthedistancemetric(Ep

7、s)istermedasEps–neighborhoodofprepresentedasNEps(p)KeyideaEverypointinaclusterhasneighborhoodofpointsforagivenradius.Itshouldcontainaminimumnumberofpointsinthatregioni.e.,thedensityintheneighborhoodshouldexceedsomethreshold.Theshapeofthisneig