ABSTRACT DATA MINING FOR A WEB-BASED EDUCATIONAL S

DATAMININGFORAWEB-BASEDEDUCATIONALSYSTEMByBehrouzMinaei-BidgoliADISSERTATIONSubmittedtoMichiganStateUniversityinpartialfulfillmentoftherequirementsforthedegreeofDOCTOROFPHILOSOPHYDepartmentofComputerScienceandEngineering2004iiABSTRACTDATAMININGFORAWEB-BASEDEDUCATIONALSYSTEMByBehrouzMinaei-BidgoliWeb-basededucationaltechnologiesalloweducatorstostudyhowstudentslearn(descriptivestudies)andwhichlearningstrategiesaremosteffective(causal/predictivestudies).Sinceweb-basededucationalsystemsarecapableofcollectingvastamountsofstudentprofiledata,dataminingandknowledgediscoverytechniquescanbeappliedtofindinterestingrelationshipsbetweenattributesofstudents,assessments,andthesolutionstrategiesadoptedbystudents.Thefocusofthisdissertationisthree-fold:1)tointroduceanapproachforpredictingstudentperformance;2)touseclusteringensemblestobuildanoptimalframeworkforclusteringweb-basedassessmentresources;and3)toproposeaframeworkforthediscoveryofinterestingassociationruleswithinaweb-basededucationalsystem.Takentogetherandusedwithintheonlineeducationalsetting,thevalueofthesetasksliesinimprovingstudentperformanceandtheeffectivedesignoftheonlinecourses.First,thisresearchpresentsanapproachtoclassifyingstudentcharacteristicsinordertopredictperformanceonassessmentsbasedonfeaturesextractedfromloggeddatainaweb-basededucationalsystem.Weshowthatasignificantimprovementinclassificationperformanceisachievedbyusingacombinationofmultipleclassifiers.Furthermore,by“learning”anappropriateweightingofthefeaturesviaageneticalgorithm(GA),wehaveiiisuccessfullyimprovedtheaccuracyofthecombinedclassifierperformancebyanother10-12%.Suchclassificationisthefirststeptowardsa“recommendationsystem”thatwillprovidevaluable,individualizedfeedbacktostudents.Second,thisprojectextendsprevioustheoreticalworkregardingclusteringensembleswiththegoalofcreatinganoptimalframeworkforcategorizingweb-basededucationalresources.Weproposebothnon-adaptiveandadaptiveresamplingschemesfortheintegrationofmultipleclusterings(independentanddependent).Experimentalresultsshowimprovedstabilityandaccuracyforclusteringstructuresobtainedviabootstrapping,subsampling,andadaptivetechniques.Theseimprovementsofferinsightsintospecificassociationswithinthedatasets.Finally,thisstudyturnstowarddevelopingatechniquefordiscoveringinterestingassociationsbetweenstudentattributes,problemattributes,andsolutionstrategies.Weproposeanalgorithmforthediscoveryof“interesting”associationruleswithinaweb-basededucationalsystem.Themainfocusisonmininginterestingcontrastrules,whicharesetsofconjunctiverulesdescribinginterestingcharacteristicsofdifferentsegmentswithinapopulation.Inthecontextofweb-basededucationalsystems,contrastruleshelptoidentifyattributescharacterizingpatternsofperformancedisparitybetweenvariousgroupsofstudents.Weproposeageneralformulationofcontrastrulesaswellasaframeworkforfindingsuchpatterns.Examiningthesecontrastscanimprovetheonlineeducationalsystemsforbothteachersandstudents–allowingformoreaccurateassessmentandmoreeffectiveevaluationofthelearningprocess.ivThisdissertationisdedicatedto:myparentsmywife,mysonMohsen,mydaughtersMaryamandZahra,andtowhoeverservetheTruthfortheTruthitself.vAcknowledgementsItiswithmuchappreciationandgratitudethatIthankthefollowingindividualswhodedicatedthemselvestothesuccessfulcompletionofthisdissertation.Dr.BillPunch,mymajorprofessorandadvisor,maintainedhispostbymysidefromtheinceptiontothecompletionofthisresearchproject.Hegavemeguidancethroughoutmyresearch.Withouthisknowledge,patience,andsupportthisdissertationwouldhavenotbeenpossible.TheknowledgeandfriendshipIgainedfromhimwilldefinitelyinfluencetherestofmylife.Otherteachershaveinfluencedmythinkingandguidedmywork.IwouldalsothankprofessorAnilJain.Itwasmyhonortobeastudentinhispatternrecognitionclasses,whichdefinedmyinterestinthefieldformanyyearstocome.Hisprofessionalguidancehasbeenasourceofinspirationandvitalinthecompletionofthiswork.IowemanyinspirationalideastoDr.Pang-NingTanwhosekeeninsightsandvaluablediscussionsoftengavemestimulatingideasinresearch.Thoughkeptbusybyhiswork,heisalwayswillingtosharehistimeandknowledgewithme.Forthesereasons,IwishhewouldhavebeenatMichiganStateUniversitywhenIbeganmyresearch.IalsoowemanythankstoDr.GerdKortemeyerforhisextraordinarysupportandpatience.Hisproductivesuggestionsandourdiscussionscontributedenormouslytothiswork.FromthefirsttimeIsteppedthroughthedoorofLiteLabintheDivisionofScienceandMathematicsEducation,untilnow,Gerdhasallowedmefreedomtoexploremyowndirectionsinresearch.Hesupportedmemateriallyandmorallyformanyyears,andforthatIamverygrateful.IamgratefultoDr.Esfahanianfortakingtimetoserveonvitheguidancecommitteeandoverseeingthiswork.Ideeplyappreciatehissupportandinsightfulsuggestions.HisgreatorganizationofthegraduateofficeintheComputerScienceDepartmenthelpsadvancegraduateresearch.Ihavelearnedmuchfromhim,bothasateacherandasafriend.IamalsogratefultomycolleaguesintheGARAGeLaboratoryespeciallyAlexanderTopchy,andallmyfriendsintheLON-CAPAdevelopersgroup:HelenKeefe,FeliciaBerryman,StuartRaeburn,andAlexa