早期心脏病预测基于神经模糊系统的研制(IJITCS-V4-N9-3)

I.J.InformationTechnologyandComputerScience,2012,9,22-28PublishedOnlineAugust2012inMECS()DOI:10.5815/ijitcs.2012.09.03Copyright©2012MECSI.J.InformationTechnologyandComputerScience,2012,9,22-28DevelopmentofNeuro-fuzzySystemforEarlyPredictionofHeartAttackObanijesuOpeyemiDepartmentofComputerScienceandEngineering,[e-mail]terScienceandEngineering,[e-mail]—Thisworkisaimedatprovidinganeuro-fuzzysystemforheartattackdetection.Theneuro-fuzzysystemwasdesignedwitheightinputfieldandoneoutputfield.Theinputvariablesareheartrate,exercise,bloodpressure,age,cholesterol,chestpaintype,bloodsugarandsex.Theoutputdetectstherisklevelsofpatientswhichareclassifiedinto4differentfields:verylow,low,highandveryhigh.Thedatasetusedwasextractedfromthedatabaseandmodeledinordertomakeitappropriateforthetraining,thentheinitialFISstructurewasgenerated,thenetworkwastrainedwiththesetoftrainingdataafterwhichitwastestedandvalidatedwiththesetoftestingdata.Theoutputofthesystemwasdesignedinawaythatthepatientcanuseitpersonally.Thepatientjustneedtosupplysomevalueswhichserveasinputtothesystemandbasedonthevaluessuppliedthesystemwillbeabletopredicttherisklevelofthepatient.IndexTerms—ANFIS,AdaptativeNeuro-FuzzySystem,Fuzzification,MembershipFunction,FuzzyRule,MembershipFunctionI.IntroductionHeartandbloodvesseldiseasescalledcardiovasculardiseasesincludenumerousproblems,manyofwhicharerelatedtoaprocesscalledatherosclerosis.Atherosclerosisisaconditionthatdevelopswhenasubstancecalledplaquebuildsupinthewallsofthearteries.Thisbuild-upnarrowsthearteriesmakingitharderforbloodtoflowthrough.Thiscancausebloodclotformationwhichcancauseaheartattack(alsocalledMyocardialInfarction)orstroke[1].Whenaheartattackoccurs,thespeedofdetectionandquickinterventionishighlyessentialtosavethelifeofheartattackpatientandtopreventheartdamage.Nowadays,theuseofcomputertechnologyinthefieldofmedicinehashighlyincreased[2].Theuseofintelligentsystemssuchasneuralnetwork,fuzzylogic,geneticalgorithmandneuro-fuzzysystemshashighlyhelpedincomplexanduncertainmedicaltaskssuchasdiagnosisofdiseases[3].Overthelastfewdecades,neuralnetworksandfuzzysystemshaveestablishedtheirreputationasalternativeapproachestointelligentinformationprocessingsystems.Bothhavecertainadvantagesoverclassicalmethods,especiallywhenvaguedataorpriorknowledgeisinvolved.However,theirapplicabilitysufferedfromseveralweaknessesoftheindividualmodels.Therefore,combinationsofneuralnetworkswithfuzzysystemshavebeenproposed,wherebothmodelscomplementeachother.Neuro-fuzzyhybridizationresultsinahybridintelligentsystemthatsynergizesthesetwotechniquesbycombiningthehuman-likereasoningstyleoffuzzysystemswiththelearningandconnectioniststructureofneuralnetworks[4].Thebasicideaofcombiningfuzzysystemsandneuralnetworksistodesignanarchitecturethatusesafuzzysystemtorepresentknowledgeinaninterpretablemannerandthelearningabilityofaneuralnetworktooptimizeitsparameters[5].Theremainingsectionsofthepaperareorganizedasfollows:InSection2thearchitectureofanAdaptativeNeuro-fuzzyInferenceSystem(ANFIS)waspresented.Thedesignoftheneuro-fuzzymodelforheartattackdetectionwaspresentedinSection3.Section4describesthedatasetused.TheexperimentalresultsaredescribedinSection5.Theconclusionoftheworkisgiveninsection6.II.AdaptativeNeuro-FuzzyInferenceSystemAdaptativeNeuro-FuzzyInferenceSystem(ANFIS)isoneofthehybridneuro-fuzzyinferenceexpertsystemsanditworksinTakagi-Sugeno-typefuzzyinferencesystem,whichwasdevelopedbyJyh-ShingandRogerJang1993[6].Thetechniqueprovideamethodforthefuzzymodelingproceduretolearninformationaboutadataset,inordertocomputethemembershipfunctionparametersthatbestallowtheassociatedfuzzyinferencesystemtotrackthegiveninput/outputdata.Thislearningmethodworksinamannersimilartothatofneuralnetworks.A.ANFISArchitectureDevelopmentofNeuro-fuzzySystemforEarlyPredictionofHeartAttack23Copyright©2012MECSI.J.InformationTechnologyandComputerScience,2012,9,22-28TheANFISarchitectureusestwofuzzyif-thenrulesbasedonafirstorderSugenomodel:whereandaretheinputs,andarethefuzzysets,aretheoutputswithinthefuzzyregionspecifiedbythefuzzyrule,,andarethedesignparametersthataredeterminedduringthetrainingprocess.Thearchitectureconsistsoffivelayersofnodes.Outofthefivelayers,thefirstandthefourthlayersconsistofadaptivenodeswhilethesecond,thirdandfifthlayersconsistoffixednodes[7].Theadaptivenodesareassociatedwiththeirrespectiveparameters,getdulyupdatedwithsubsequentiterationwhilethefixednodesaredevoidofanyparameters.Figure1showstheANFISarchitectureFig1ANFISARCHITECTURELayer1:Fuzzificationlayer:Everynodeinthelayer1isanadaptivenode.Theoutputsoflayer1arethefuzzymembershipgradeoftheinputs,whicharegivenby:(x),(1)(y),(2)wherexandyaretheinputstonodei,whereAisalinguisticlabel(small,large)associatedwiththisnodefunctionisthemembershipfunctionofAianditspecifiesthedegreetowhichthegivenx,ysatisfiesthequantifierAi.(x),(y)canadoptanyfuzzymembershi