基于机器学习技术的泌尿系统疾病诊断(IJISA-V7-N5-1)

I.J.IntelligentSystemsandApplications,2015,05,1-7PublishedOnlineApril2015inMECS()DOI:10.5815/ijisa.2015.05.01Copyright©2015MECSI.J.IntelligentSystemsandApplications,2015,05,1-7UrinarySystemDiseasesDiagnosisUsingMachineLearningTechniquesSeyyidAhmedMedjahedUniversityofSciencesandTechnologyMohamedBoudiafUSTO-MB,FacultyofMathematicsandComputerScience,Oran,31000,AlgeriaEmail:[e-mail]ersityofHave,Havre,76600,FranceEmail:[e-mail]versityofSciencesandTechnologyMohamedBoudiafUSTO-MB,FacultyofMathematicsandComputerScience,Oran,31000,AlgeriaEmail:[e-mail]—Theurinarysystemistheorgansystemresponsiblefortheproduction,storageandeliminationofurine.Thissystemincludeskidneys,bladder,uretersandurethra.Itrepresentsthemajorsystemwhichfiltersthebloodandanyimbalanceofthisorgancanincreasestherateofbeinginfectedwithdiseases.TheaimofthispaperistoevaluatetheperformanceofdifferentvariantsofSupportVectorMachinesandk-NearestNeighborwithdifferentdistancesandtrytoachieveasatisfactoryrateofdiagnosis(infectedornon-infectedurinarysystem).Weconsiderbothdiseasesthataffecttheurinarysystem:inflammationofurinarybladderandnephritisofrenalpelvisorigin.Ourexperimentationwillbeconductedonthedatabase“AcuteInflammationsDataSet”obtainedfromUCIMachineLearningRepository.Weusethefollowingmeasurestoevaluatetheresults:classificationaccuracyrate,classificationtime,sensitivity,specificity,positiveandnegativepredictivevalues.IndexTerms—UrinarySystem,Diagnosis,SupportVectorMachine,k-NearestNeighbor,Distance.I.INTRODUCTIONMachinelearningisthescientificdisciplineconcernedwiththedevelopment,analysisandimplementationofautomatedmethodsthatallowtoamachinetoevolvethroughalearningprocess,andsofulfillthetasksthataredifficultorimpossibletofillbymoreconventionalalgorithmicmeans.Learningalgorithmscanbecategorizedaccordingtothelearningmode:supervisedlearning,unsupervisedlearningandsemi-supervisedlearning.Supervisedlearningiswidelyusedinthemedicalfieldasanensembleofmethodsforthemedicaldiagnosistohelpthedoctorandtogetabetterdiagnosis.Amongthesemethodsare:SupportVectorMachines,NeuronNetwork,k-NearestNeighbor,etc.Manyresearchershavedevelopedexpertsystemstosolvecomplexproblemsofmedicaldiagnosis(suchascancerdiagnosis[1],[2],acuteinflammationinurinarysystemdiagnosis[3],[4],etc.)byreasoningaboutknowledgeandalsobyusingdifferentlearningmethods.Supportvectormachineandk-nearestneighborhavebeenwidelyusedinmedicaldiagnosisfield.In[5],theauthorshaveusedSVMwithRBFkerneltodiagnosistheheartdisease.Leungetal.[6],havedevelopedadataminingframeworktodiagnosistheHepatitisB.In[7],theauthorshaveproposedthreeneuralnetworkapproachesandappliedinhepatitisdiseases.A.Kharratetal.[8],proposedanovelvariantofSupportvectormachinecalledEvolutionarySVMformedicaldiagnostic.Inthispaper,thefirstworkistoanalysisandtoevaluatetheperformanceofdifferentvariantofSupportVectorMachinesandk-NearestNeighboralgorithmwithdifferentdistances,inthecontextofthediagnosisofacuteinflammationinurinarysystem(infectedornon-infectedwithinflammationofurinarybladderornephritisofrenalpelvisorigin).Thesecondworkistoreachahighclassificationaccuracyrate.Theevaluationsofperformancehavebeenconductedintermof:classificationaccuracyrate,classificationtime,sensitivity,specificity,positiveandnegativepredictivevalues.Thepaperisorganizedasfollows:FirstwerecallsomedefinitionofSupportVectorMachinesandthedifferentapproachesusinginthisstudy.InSec.IV,wepresentthek-NearestNeighborandthedifferentdistance.InSec.V,weanalyzetheresultofthedifferentmethodsandfinallyweconcludewithsomeperspectives.II.SUPPORTVECTORMACHINESSupportVectorMachinewasintroducedbyVladimirN.Vapnikin1995[9],[10]anditbecomesratherpopularsince.SVMperformsclassificationbyconstructingahyperplanethatoptimallyseparatesthedataintotwocategories(inthecaseofbinaryclassification).ThemodelsofSVMarecloselyrelatedtoNeuralNetworks.2UrinarySystemDiseasesDiagnosisUsingMachineLearningTechniquesCopyright©2015MECSI.J.IntelligentSystemsandApplications,2015,05,1-7SVMworkswellinpracticeandhasbeenusedacrossawiderangeofapplicationsfromrecognizedhand-writtendigits,faceidentification,bioinformatics,etc.ThegoalofSVMistofindtheoptimalhyperplanethatseparatesclustersofvectorinsuchawaythatcaseswithonecategoryofthetargetvariableareononesideoftheplaneandcaseswiththeothercategoryareontheothersizeoftheplane[11].Thevectorsnearthehyperplanearethesupportvectors.Findtheoptimalhyperplaneisequivalenttoreformulatetheclassificationproblemtoanoptimizationproblem[12].Fig.1.TheoptimalHyperplane.A.MathematicalformulationConsiderabinaryclassificationproblemwithNtrainingpointsiiyx,,Ni,...,1,whereeachinputixhasDattributesandisoneoftwoclasses1,1iy.Thedatatraininghavethefollowingform:iiyx,whereDiiRxyNi,1,1,,...,1(1)Thehyperplanecanbedescribedby:0,bxwi(2)wherewisthenormtothehyperplaneandwbistheperpendiculardistancefromthehyperplanetotheorigin.Findingtheoptimalhyperplaneisequivalenttosolvingthefol