基于支持向量機(jī)的外顯子與內(nèi)含子識(shí)別及回歸預(yù)測(cè)研究.pdf

1、湖南農(nóng)業(yè)大學(xué)碩士學(xué)位論文基于支持向量機(jī)的外顯子與內(nèi)含子識(shí)別及回歸預(yù)測(cè)研究姓名：張永生申請(qǐng)學(xué)位級(jí)別：碩士專業(yè)：農(nóng)業(yè)昆蟲與害蟲防治指導(dǎo)教師：袁哲明20070601AbstractSupportvectormachine(SVM)includesupposevectorclassification(SVC)andsupportvectormachineregression(SVR)，whichwasputforwardin1995anddev

2、elopsmostrapidlyamongallofmachinelearningmethodsSVMisbasedontheprincipleofstructuralriskminimization，andthenresolvessuchpracticalproblemsasnonlinearityoverfit，curseofdimensionality，localminima，smallsampleslearningandhash

3、ighgeneralizationThispaperstudiestherecognitionofexonandintron，multidimensionaltimeseriespredictioninlongitudinaldataandquantitativestructure—activityrelationshipinnonlongitudinaldambyimprovingSVMThemaincontentsandresult

4、sareasfollows：1)AnovelmethodoffeatureextractionnamedmultiscalecomponentandcorrelationWasproposedandWasappliedtotherecognitionofexonandintronbasedonFisherdiscriminantorSVC1Andtheexonsensitivity‘鑫)，exonspecificity(品)，intro

5、nsensitivity@)andcorrelationcoefficient(cofromtestsetsare09240，09893，O9900andO9160respectivelyThemethodhadtheadvantagesofsimplealgorithm，hi曲accuracy，wideusage，etc2)BasedonSVRandcontrolledautoregressive(CAR)，weproposedane

6、wnonlinearmultidimensionaltimeseriesmethodnamedSVRCARthatcaIlshowthedynamiccharacteristicsofsamplesetaswallastheeffectofenvironmentalfactorsToevaluatetheperformanceofSVRCA&wecompareditspredictionswiththoseoffourothercomm

7、onlyusedmethods，usingtwosetsoflongitudinaldataandone—steppredictionTheresultsshowedthatSVR—CARhadthehi曲estaccuracyinpredictionamongthefivemethods。SVRCARhasthepotentialtobewidelyusedforpredictionsinvolvingmultidimensional

8、timeseriesdatainecolo現(xiàn)agriculturalsciencesandeconomics3)Toimprovethepredicationprecisioninquantitativestructureactivityrelationship(QSAR)anovelnonlinearcombinatorialforecastmethodbasedonSVRand￡nearneighborgroupWaspropose

9、dThepredictionresultsofnonlongitudinaldataofQSARforsubstitutedanilinesandphenolstODaphniamagnaStrausshowedthatthenovelcombinationmethodhadthehighestpredictionprecisionamongthetenmethodsandcharacterizedthenonlinearrelatio