模糊和神经模糊系统的性能改进：学龄儿童学习障碍的预测(IJISA-V5-N12-3)

I.J.IntelligentSystemsandApplications,2013,12,34-52PublishedOnlineNovember2013inMECS()DOI:10.5815/ijisa.2013.12.03Copyright©2013MECSI.J.IntelligentSystemsandApplications,2013,12,34-52PerformanceImprovementofFuzzyandNeuroFuzzySystems:PredictionofLearningDisabilitiesinSchool-ageChildrenJulieM.DavidDept.ofComputerApplications,MESCollege,Marampally,Aluva,Cochin-683107,IndiaE-mail:[e-mail],CochinUniversityofScience&Technology,Cochin-682022,IndiaE-mail:[e-mail]—LearningDisability(LD)isaclassificationincludingseveraldisordersinwhichachildhasdifficultyinlearninginatypicalmanner,usuallycausedbyanunknownfactororfactors.LDaffectsabout15%ofchildrenenrolledinschools.ThepredictionoflearningdisabilityisacomplicatedtasksincetheidentificationofLDfromdiversefeaturesorsignsisacomplicatedproblem.Thereisnocureforlearningdisabilitiesandtheyarelife-long.Theproblemsofchildrenwithspecificlearningdisabilitieshavebeenacauseofconcerntoparentsandteachersforsometime.Theaimofthispaperistodevelopanewalgorithmforimputingmissingvaluesandtodeterminethesignificanceofthemissingvalueimputationmethodanddimensionalityreductionmethodintheperformanceoffuzzyandneurofuzzyclassifierswithspecificemphasisonpredictionoflearningdisabilitiesinschoolagechildren.InthebasicassessmentmethodforpredictionofLD,checklistsaregenerallyusedandthedatacasesthuscollectedfullydependsonthemoodofchildrenandmayhavealsocontainredundantaswellasmissingvalues.Therefore,inthisstudy,weareproposinganewalgorithm,viz.thecorrelationbasednewalgorithmforimputingthemissingvaluesandPrincipalComponentAnalysis(PCA)forreducingtheirrelevantattributes.Afterthestudy,itisfoundthat,thepreprocessingmethodsappliedbyusimprovesthequalityofdataandtherebyincreasestheaccuracyoftheclassifiers.ThesystemisimplementedinMathworksSoftwareMatLab7.10.Theresultsobtainedfromthisstudyhaveillustratedthatthedevelopedmissingvalueimputationmethodisverygoodcontributioninpredictionsystemandiscapableofimprovingtheperformanceofaclassifier.IndexTerms—ANFIS,DataMining,FIS,LearningDisability,MembershipFunctionI.IntroductionThetermDataMiningorKnowledgeDiscoveryinDatabases,hasbeenadoptedforafieldofresearchdealingwiththeautomaticdiscoveryofimplicitinformationorknowledgewithindatabases[1,2].Dataminingisacollectionoftechniquesforefficientautomateddiscoveryofpreviouslyunknown,valid,novel,usefulandunderstandablepatternsinlargedatabases.Accordingtoawidelyacceptedformaldefinitiongivensubsequently,dataminingisthenon-trivialextractionofimplicitpreviouslyunknownandpotentiallyusefulinformationaboutdata[3].Inrecentyearsthesizesofdatabaseshasincreasedrapidly.Thishasledtoagrowinginterestinthedevelopmentoftoolscapableintheautomaticextractionofknowledgefromdata.Diversefieldssuchasmarketing,customerrelationshipmanagement,engineering,medicine,crimeanalysis,expertprediction,webminingandmobilecomputingbesidesothersutilizedatamining[4].Amajorityofareasrelatedtomedicalservicessuchaspredictionofeffectivenessofsurgicalprocedures,medicaltests,medicationandthediscoveryofrelationshipamongclinicalanddiagnosisdataalsomakeuseofdataminingmethodologies[5].Learningdisabilityisageneraltermthatdescribesspecifickindsoflearningproblems.Itisaneurologicalconditionthataffectsachild'sbrainandimpairshisabilitytocarryoutoneormanyspecifictasks[6].Thelearningdisabilityaffectedchildrenareneitherslownormentallyretarded[7].Anaffectedchildcanhavenormaloraboveaverageintelligence.Theymayhavedifficultypayingattention,withreadingorletterrecognition,orwithmathematics.Itdoesnotmeanthatchildrenwhohavelearningdisabilitiesarelessintelligent.Infact,manychildrenwhohavelearningdisabilitiesaremoreintelligentthantheaveragechild.Mentalretardation,emotionaldisordersandpoorsocioeconomicstatusarenotconsideredaslearningdisabilities.ThemainthingtorememberwhenlearningdisabilitiesarediscussedisthatwithproperPerformanceImprovementofFuzzyandNeuroFuzzy35Systems:PredictionofLearningDisabilitiesinSchool-ageChildrenCopyright©2013MECSI.J.IntelligentSystemsandApplications,2013,12,34-52intervention,teachingandlearningtechniques,achildwithspecialneedsrelatedtooneofthesedisabilitiescansucceedinschool.Thisiswhyachildwithalearningdisabilityisoftenwronglylabelledasbeingsmartbutlazy.Thelearningdisabilitycancauseachildtohavetroubleinlearningandusingcertainskills.Theskillsmostoftenaffectedare:reading,writing,listening,speaking,reasoninganddoingmath[6].Learningdisabilitiesvaryfromchildtochild.OnechildwithLDmaynothavethesamekindoflearningproblemsasanotherchildwithLD.Thereisnocureforlearningdisabilities[5,8].Theyarelife-long.However,childrenwithLDcanbehighachieversandcanbetaughtwaystogetaroundthelearningdisability.Withtherighthelp,childrenwithLDcananddolearnsuccessfully.Missingattributevaluescommonlyexistinrealworlddataset.Theymayoccurfromverybeginningstageviz.datacollectingprocessandduringredundantdiagnosetests,unknowndata,etc.Discardingalldatacontainingthemissingattributevaluescannotfullypreservethecharac