Toacceleratethetrainingofsupportvectordomaindescription(SVDD),confidencesupportvectordomaindescription(CSVDD)isproposedbasedontheobservationthatthedescriptionboundaryisdeterminedbyasmallsubsetoftrainingdatacalledsupportvectors.Namely,thenumberoftrainingsamplesintheuserdefinedsphereiscalculatedandtakenastheconfidencemeasure,accordingtowhichthetrainingsamplesarerankedinascendingorder.ThoseformerrankedonesareselectedastheboundarytargetsfortheSVDDtraining.SimulationsonUCIdatademonstratetheeffectivenessandsuperiorityofCSVDD:thenumberoftrainingtargetsandthetrainingtimearereducedwithoutanylossofaccuracy.
