统一聚类与分类模型提高学生就业能力预测(IJISA-V9-N9-2)

I.J.IntelligentSystemsandApplications,2017,9,10-18PublishedOnlineSeptember2017inMECS()DOI:10.5815/ijisa.2017.09.02Copyright©2017MECSI.J.IntelligentSystemsandApplications,2017,9,10-18AUnifiedModelofClusteringandClassificationtoImproveStudents’EmployabilityPredictionPoojaThakarBanasthaliUniversity,Jaipur,304022,IndiaE-mail:[e-mail]asthan,Jaipur,304022,IndiaE-mail:[e-mail]ity,Jaipur,304022,IndiaE-mail:[e-mail]:02May2017;Accepted:24July2017;Published:08September2017Abstract—DataMiningisgainingimmensepopularityinthefieldofeducationduetoitspredictivecapabilities.But,mostoftheprioreffortinthisareaisonlydirectedtowardspredictionofperformanceinacademicresultsonly.Nowadays,educationhasbecomeemploymentoriented.Verylittleattemptismadetopredictstudents’employability.Precisepredictionofstudents’performanceincampusplacementsatanearlystagecanidentifystudents,whoareattheriskofunemploymentandproactiveactionscanbetakentoimprovetheirperformance.Existingresearchesonstudents’employabilitypredictionareeitherbasedupononlyonetypeofcourseoronsingleUniversity/Institute;thusisnotscalablefromonecontexttoanother.Withthisnecessity,theconceptionofaunifiedmodelofclusteringandclassificationisproposedinthispaper.Withthenotionofunification,dataofprofessionalcoursesnamelyEngineeringandMastersinComputerApplicationsstudentsarecollectedfromvariousuniversitiesandinstitutionspanIndia.Dataislarge,multivariate,incomplete,heterogeneousandunbalancedinnature.Todealwithsuchadata,aunifiedpredictivemodelisbuiltbyintegratingclusteringandclassificationtechniques.Two-Levelclustering(k-meanskernel)withchi-squareanalysisisappliedatthepre-processingstagefortheautomatedselectionofrelevantattributesandthenensemblevoteclassificationtechniquewithacombinationoffourclassifiersnamelyk-star,randomtree,simplecartandtherandomforestisappliedtopredictstudents’employability.Proposedframeworkprovidesageneralizedsolutionforstudentemployabilityprediction.Comparativeresultsclearlydepictmodelperformanceovervariousclassificationtechniques.Also,whentheproposedmodelisapplieduptothelevelofthestate,classificationaccuracytouches96.78%and0.937kappavalue.IndexTerms—Clustering,Classification,DataMining,Employability,Prediction,EducationI.INTRODUCTIONThelatestreportpublishedbyIndiaTodayonMarch20,2017,revealedthatoutofeightlakhengineeringgraduatesinthecountry,60%remainunemployed[1].InasimilarreportlastyearonJuly13,2016,itpublishedthatonly7%ofengineersaresuitableforcoredomainjobs[2].Theneedofthehouristoaccuratelypredictgraduatesemployabilityintheveryfirstyearoftheircourseenrollmentsothatmoreeffectualactionsandpoliciescanbeimplementedontime.Institutesmaintaindetailsofstudentsfromtheirenrollmenttilltheypassout,whichincludesacademicrecords,personaldetails,andvariousskillbasedtestrecordssuchasaptitudetestandpsychometrictest.Thisdatacanbeofimmenseuseifutilizedforanalysis.Studiesreflectthepotentialofsuchdatatoguidestudentsforbetteremployability.But,suchdatasuffersfromtwomajorinherentproblemsi.e.unbalancedandmultivariate[3].Moreover,mostoftheexistingresearcheseitherconsideradatasetofonlyonetypeofcourseoroneUniversity/Instituteforpredictiveanalysisofstudents’employability[4][5].EducationalDecisionSystemsarethusextremelycustomizedtofulfilltheneedsofthespecificInstitute.Thereisnounifiedapproachthatcanbeusedacrossinstituteswithanytypeofdataset.Thispaperpresentsaunifiedmodelthatservestwomajorobjectives.A.Toautomatetheselectionofrelevantattributesfromthesetofthemultivariatedataset(151attributesinthepresentstudy)atpreprocessingstage.Proposedmodelusestwo-levelclusteringtoreducethedatasetautomaticallybaseduponAUnifiedModelofClusteringandClassificationtoImproveStudents’EmployabilityPrediction11Copyright©2017MECSI.J.IntelligentSystemsandApplications,2017,9,10-18classificationresults.Finally,findingtherelevantsetofattributeswithchi-squareanalysis.B.Toconstructtheunifiedpredictionmodelbaseduponmostsuitableclassificationalgorithm(s).ProposedModelintegratesfourbestclassifierswithvoteensemblemethodtopredictstudent’semployability(i.e.placedorunplacedinon-campusplacementdrives).Researcherssuggestthatinitialmultidimensionaldatasetshouldbeputintoself-learningmodetogeneratehomogeneousgroups[6].Itisalsoprovedthatclusteringappliedonattributessetatpre-processingstagehelpsinparsimoniousselectionofvariablesandimprovestheperformanceofpredictivealgorithms[3].EnsembleModelenhancestheclassificationaccuracybyintegratingthepredictionaccuracyofbaseclassifiers.Multipleclassifierbasedsystemsincreasetheperformanceofindividualclassifiers[7].Thus,proposedmodelnotonlyoutperformsthepredictionperformanceofvariousclassifiersbutalsohelpinfindingtherelevantnumberofattributesautomatically.Restofthepaperdescribestheproposedmodelanditsresultsindetails.SectionIIprovidesadetailedreviewofexistingliteratureinthedomainofstudents’employabilityprediction.SectionIIIdescribestheexperimentalsettingoftoolsandtechniquesused.