GAY-MLP神经网络：糖尿病诊断的混合智能系统(IJISA-V8-N1-6)

I.J.IntelligentSystemsandApplications,2016,1,49-59PublishedOnlineJanuary2016inMECS()DOI:10.5815/ijisa.2016.01.06Copyright©2016MECSI.J.IntelligentSystemsandApplications,2016,1,49-59GA_MLPNN:AHybridIntelligentSystemforDiabetesDiseaseDiagnosisDilipKumarChoubeyBirlaInstituteofTechnology,ComputerScience&Engineering,Mesra,Ranchi,IndiaEmail:[e-mail]tuteofTechnology,ComputerScience&Engineering,Mesra,Ranchi,IndiaEmail:[e-mail]—Diabetesisaconditioninwhichtheamountofsugarinthebloodishigherthannormal.Classificationsystemshavebeenwidelyusedinmedicaldomaintoexplorepatient’sdataandextractapredictivemodelorsetofrules.Theprimeobjectiveofthisresearchworkistofacilitateabetterdiagnosis(classification)ofdiabetesdisease.Therearealreadyseveralmethodologywhichhavebeenimplementedonclassificationforthediabetesdisease.Theproposedmethodologyimplementedworkin2stages:(a)InthefirststageGeneticAlgorithm(GA)hasbeenusedasafeatureselectiononPimaIndianDiabetesDataset.(b)Inthesecondstage,MultilayerPerceptronNeuralNetwork(MLPNN)hasbeenusedfortheclassificationontheselectedfeature.GAisnotedtoreducenotonlythecostandcomputationtimeofthediagnosticprocess,buttheproposedapproachalsoimprovedtheaccuracyofclassification.Theexperimentalresultsobtainedclassificationaccuracy(79.1304%)andROC(0.842)showthatGAandMLPNNcanbesuccessfullyusedforthediagnosingofdiabetesdisease.IndexTerms—PimaIndianDiabetesDataset,GA,MLPNN,DiabetesDiseaseDiagnosis,FeatureSelection,Classification.I.INTRODUCTIONDiabetesisachronicdiseaseandamajorpublichealthchallengeworldwide.Diabeteshappenswhenabodyisnotabletoproduceorrespondproperlytoinsulin,whichisneededtomaintaintherateofglucose.Diabetescanbecontrolledwiththehelpofinsulininjections,acontrolleddiet(changingeatinghabits)andexerciseprograms,butnowholecureisavailable.Diabetesleadstomanyotherdiseasesuchasblindness,bloodpressure,heartdisease,kidneydiseaseandnervedamage[15].Main3diabetessignsare:-Increasedneedtourinate(Polyuria),Increasedhunger(Polyphagia),Increasedthirst(Polydipsia).Therearetwomaintypesofdiabetes:Type1(JuvenileorInsulinDependentorBrittleorSugar)DiabetesandType2(AdultonsetorNonInsulinDependent)Diabetes.Type1Diabetesmostlyhappenstochildrenandyoungadultsbutcanaffectatanyage.Forthistypeofdiabetes,betacellsaredestructedandpeoplesufferingfromtheconditionrequireinsulininjectionregularlytosurvive.Type2Diabetesisthemostcommontypeofdiabetes,inwhichpeoplearesufferingatleast90%ofallthediabetescases.Thistypemostlyhappenstothepeoplemorethanfortyyearsoldbutcanalsobefoundinyoungerclasses.Inthistype,bodybecomesresistanttoinsulinanddoesnoteffectivelyusetheinsulinbeingproduced.Itcanbecontrolledwithlifestylemodification(ahealthydietplan,doingexerciseregularly),oralmedications(takingtablets).Insomeextremecases,insulininjectionsmayalsoberequiredbutnowholecurefordiabetesisavailable.Inthispaper,GAhasbeenusedasaFeatureselectioninwhichamong8attributes,4attributeshavebeenselected.ThemainpurposeofFeatureselectionistoreducethenumberoffeaturesusedinclassificationwhilemaintainingacceptableclassificationaccuracyandROC.Limitingthenumberoffeatures(dimensionality)isimportantinstatisticallearning.WiththehelpofFeatureselectionprocesswecansaveStoragecapacity,Computationtime(shortertrainingtimeandtesttime),ComputationcostandincreasesClassificationrate,Comprehensibility.MLPNNaresupervisedlearningmethodforclassification.Here,MLPNNhavebeenusedfortheclassificationoftheDiabetesdiseasediagnosis.Therestofthepaperisorganizedasfollows:BriefdescriptionofGAandMLPNNareinsectionII,RelatedworkispresentedinsectionIII,ProposedmethodologyisdiscussedinsectionIV,ResultsandDiscussionaredevotedtosectionV,ConclusionandFutureDirectionarediscussedinsectionVI.II.BRIEFDESCRIPTIONOFGAANDMLPNNA.GAJohnHollandintroducedgeneticAlgorithmGAinthe1970atUniversityofMichigan(US).GAisanadaptivepopulationbasedoptimizationtechnique,whichisinspiredbyDarwin’stheory[10]aboutsurvivalofthefittest.GAmimicsthenaturalevolutionprocessgivenbytheDarwini.e.,inGAthenextpopulationis50GA_MLPNN:AHybridIntelligentSystemforDiabetesDiseaseDiagnosisCopyright©2016MECSI.J.IntelligentSystemsandApplications,2016,1,49-59evolvedthroughsimulatingoperatorsofselection,crossoverandmutation.JohnHollandisknownasthefatheroftheoriginalgeneticalgorithmwhofirstintroducedtheseoperatorsin[16].Goldberg[13]andMichalewicz[18]laterimprovedtheseoperators.TheadvantagesinGA[17]arethatConceptseasytounderstand,solvesproblemswithmultiplesolutions,globalsearchmethods,blindsearchmethods,Gascanbeeasilyusedinparallelmachines,etc.andthelimitationareCertainoptimizationproblems,noabsoluteassuranceforaglobaloptimum,cannotassureconstantoptimizationresponsetimes,cannotfindtheexactsolution,etc.GAcanbeappliedinArtificialcreativity,Bioinformatics,chemicalkinetics,Geneexpressionprofiling,controlengineering,softwareengineering,Travelingsalesmanproblem,Mutationtesting,Qualitycontroletc.Thegeneticalgorithmusesthreemaintypesofrulesateachsteptocreatethenextgenerationfromthecurrent