自适应神经模糊学生分类(IJMECS-V4-N7-6)

I.J.ModernEducationandComputerScience,2012,7,42-49PublishedOnlineJuly2012inMECS()DOI:10.5815/ijmecs.2012.07.06Copyright©2012MECSI.J.ModernEducationandComputerScience,2012,7,42-49StudentsClassificationWithAdaptiveNeuroFuzzyMohammadSaberIrajiDepartmentofComputerscience,YoungResearchersClubsariBranch,IslamicAzadUniversity,sari,IranE-mail:[e-mail]mentofComputerEngineering,IslamicAzadUniversity,SariBranch,Sari,IranE-mail:[e-mail]omputerEngineering,AliabadKatoulBranch,IslamicAzadUniversity,AliabadKatoul,IranE-mail:[e-mail]ofComputerEngineering,AliabadKatoulBranch,IslamicAzadUniversity,AliabadKatoul,IranE-mail:[e-mail]—Identifyingexceptionalstudentsforscholarshipsisanessentialpartoftheadmissionsprocessinundergraduateandpostgraduateinstitutions,andidentifyingweakstudentswhoarelikelytofailisalsoimportantforallocatinglimitedtutoringresources.Inthisarticle,wehavetriedtodesignanintelligentsystemwhichcanseparateandclassifystudentaccordingtolearningfactorandperformance.asystemisproposedthroughLvqnetworksmethods,anfismethodtoseparatethesestudentonlearningfactor.Inourproposedsystem,adaptivefuzzyneuralnetwork(anfis)haslesserrorandcanbeusedasaneffectivealternativesystemforclassifyingstudents.IndexTerms—Adaptiveneurofuzzy,Neuralnetwork,Studentsclassification,LvqI.INTRODUCTIONPredictingstudents‟academicperformanceiscriticalforeducationalinstitutionsbecausestrategicprogramscanbeplannedinimprovingormaintainingstudents‟performanceduringtheirperiodofstudiesintheinstitutions[1].Arithmeticalandstatisticalmethodsareunabletoofferaneffectiveinferenceproceduretoperformtheevaluationoftheacademicperformancesofstudentsinamorenaturalway,usinglinguisticvariables.Thismethodmighthelpstudents,theirparents,decisionmakers,andevaluatorsinobtainingmorereliableandunderstandableresultsforastudent‟sachievement,orforagroupofstudentsandtheircomparativeevaluations.Itisimportanttopointoutthattheaimofproposedmethodisnottoreplacethetraditionalmethodofevaluation;instead,itistostrengthenthepresentsystembyprovidingadditionalinformationfordecisionmaking[2].Assessmentofthestudent‟sacademicperformance(SAP)isoneofthemostimportantpracticesusedforthreemainreasons:todecideonpassandfailureincourses,toobtainanindicationofthestudent‟sleveloflearning,andtoprovideinformationontheeffectivenessofteaching.Intraditional(statistical)methods,thestudent‟sacademicperformance(SAP)isevaluatedbasedonthemarkscollectedbyastudent.Itcanbeclassifiedintonumerouscategoriessuchassinglenumericalscoresusuallyreferringto100percent,singlelettergrades(e.g.A,B,C,D,orF),nominalscores(e.g.1,2,3...10),linguistictermssuchas„„Fail”,or„„Pass”orsinglegrade-pointsfrom0.00to4.00.Asapartofthisstudy,aweightedsumofassessmenttoolsisusedtocalculatethenumericalscoreofeachstudentasfollows:Quiz(Q)is10%,Major(M)is15%,Midterm(MD)is20%,Final(F)is40%,PerformanceAppraisals(P)is10%,andSurvey(S)is5%.Thetotaloutof100indicatesthestudent‟sacademicperformance(SAP)[2].Anumberofsocio-economic,biological,environmental,academic,andotherrelatedfactorsthatareconsideredtohaveinfluenceontheperformanceofauniversitystudentwereidentified.ThesefactorswerecarefullystudiedandharmonizedintoamanageablenumbersuitableforcomputercodingwithinthecontextoftheANNmodeling[3].Thepaperisorganizedinfivesections.AftertheintroductioninSectionІ,SectionІІ,whichalsointroducestheexistingmethodsoftheperformanceevaluation.SectionІІcontinueswithexplanationsofLvqneuralnetworkandadaptiveneuro-fuzzysystems(ANFIS)insectionІІІ.SectionІVdiscussesthefactorsaffectingonclassificationstudentbaselearning.Itcontinueswithdiscussionsonthearchitectureofhybridlearningandfuzzymodelvalidationandlvqneuralnetwork,theerrorofobservationsfortrainingdatasets.SectionVpresentsStudentsClassificationWithAdaptiveNeuroFuzzy43Copyright©2012MECSI.J.ModernEducationandComputerScience,2012,7,42-49theconclusionsoftheresearch.Thepaperendswithalistofreferences.II.LITERATUREREVIEWNeuro-fuzzysystemsareoneofthemostsuccessfulandvisibledirectionsofthateffort.Neurofuzzyhybridizationisdoneintwoways[4]:aneuralnetworkequippedwiththecapabilityofhandlingfuzzyinformation(termedfuzzyneuralnetwork)andafuzzysystemaugmentedbyneuralnetworkstoenhancesomeofitscharacteristicslikeflexibility,speed,andadapt-ability(termedneuro-fuzzysystem(NFS)orANFIS).Anadaptedneuro-fuzzysystem(NFS)isdesignedtorealizetheprocessoffuzzyreasoning,wheretheconnectionweightsofnetworkcorrespondtoparametersoffuzzyreasoning[4,5].Thesemethodologiesarethoroughlydiscussedintheliterature[4].Asecondanddistinctapproachtohybridizationisthegeneticfuzzysystems(GFSs)[6].AGFSisessentiallyafuzzysystemaugmentedbyalearningprocessbasedongeneticalgorithms(GAs).Theparameteroptimizationhasbeentheapproachusedtoadaptawiderangeofdissimilarfuzzysystems,asingeneticfuzzyclusteringorgeneticfuzzysystems[6].However,geneticfuzzysystemsarenotsubjectofthiswork.Let'srevieweducationalproductivitytheoryfirst.In1981,Wa