Outlier detection using ball descriptions with adjustable metric
David M. J. Tax (TU Delft - Multimedia Computing)
P Juszczak (TU Delft - Multimedia Computing)
E.M. Pekalska (TU Delft - Multimedia Computing)
Bob Duin (TU Delft - Multimedia Computing)
More Info
expand_more
Abstract
Sometimesnoveloroutlierdatahastobedetected.Theoutliersmayindicatesomeinterestingrareevent,ortheyshouldbedisregardedbecausetheycannotbereliablyprocessedfurther.Intheidealcasethattheobjectsarerepresentedbyverygoodfeatures,thegenuinedataformsacompactclusterandagoodoutliermeasureisthedistancetotheclustercenter.Thispaperproposesthreenewformulationsto¿ndagoodclustercentertogetherwithanoptimizedp-distancemeasure.Experimentsshowthatforsomerealworlddatasetsverygoodclassi¿cationresultsareobtainedandthat,morespeci¿cally,the1-distanceisparticularlysuitedfordatasetscontainingdiscretefeaturevalues.
No files available
Metadata only record. There are no files for this record.