摘要
Anewfuzzysupportvectormachinealgorithmwithdualmembershipvaluesbasedonspectralclusteringmethodisproposedtoovercometheshortcomingofthenormalsupportvectormachinealgorithm,whichdividesthetrainingdatasetsintotwoabsolutelyexclusiveclassesinthebinaryclassification,ignoringthepossibilityof'overlapping'regionbetweenthetwotrainingclasses.Theproposedmethodhandlessample'overlap'efficientlywithspectralclustering,overcomingthedisadvantagesofover-fittingwell,andimprovingthedataminingefficiencygreatly.Simulationprovidesclearevidencestothenewmethod.
出版日期
2012年02月12日(中国期刊网平台首次上网日期,不代表论文的发表时间)