用前馈神经网络方法评价教师的评价绩效(IJISA-V9-N3-5)

I.J.IntelligentSystemsandApplications,2017,3,34-40PublishedOnlineMarch2017inMECS()DOI:10.5815/ijisa.2017.03.05Copyright©2017MECSI.J.IntelligentSystemsandApplications,2017,3,34-40ClusteringofFacultybyEvaluatingtheirAppraisalPerformancebyusingFeedForwardNeuralNetworkApproachC.BhanuprakashDepartmentofMasterofComputerApplications,SiddagangaInstituteofTechnology,Tumkur,India.E-mail:[e-mail]omputerScienceandEngineering,SiddagangaInstituteofTechnology,Tumkur,India.E-mail:[e-mail]rofComputerApplications,SiddagangaInstituteofTechnology,Tumkur,India.E-mail:[e-mail]—Clusteringistheprocessofgroupingasetofdataobjectsintomultiplegroupsorclusterswithhighsimilaritiesanddissimilarities.DissimilaritiesandSimilaritiesareassessedontheattributevaluesdescribingtheobjectsandofteninvolvedistancemeasures.Clusteringactsasadataminingtoolbyhavingitsrootsinmanyapplicationareassuchasbiology,security,businessintelligence,websearchetc.OurInstituteiscurrentlyusingasoftwareapplicationwithaname―MeritSystem‖,whichevaluatestheperformanceofthestaffmembersregardingtheirlevelofteachingbyconsideringvariousfactors.Itcomputestheperformancelevelbycollectingfeedbackfromeverystudent.Itgivestheappraisalresultintheformof30pointsearnedtoeverystaffmember.Itactsasatoolforthemanagementofourcollegetogaugetheperformanceleveloftheteacherwhichinturnhelpstheminassessingannualincrementsandotherpromotions.ThemaindrawbackofthissystemisitsinabilityingroupingofstaffmemberslikeGroup-A,Group-B,Group-Cetc.Because,manyofthestaffmembershavescoredtheperformancepointsintherangeof21to30whichwillcreateslotofambiguitiestothemanagementtomakeclustersofstaffmemberstothesegroups.ThisissueistheprimeconcernofthispaperanditwasgivenwithanapproachtosolvethisproblembyconsideringpossibleoptimumsoftcomputingtechniquethatincludesFeedForwardNeuralNetworkapproach.IndexTerms—Clustering,FuzzyGrouping,Fuzzypartitions,Rangeofvalues,Similarities,Neuralnetworks,Hiddenlayers,feedback.I.INTRODUCTIONOnthelastweekofeverysemester,itismandatorytoeverystudenttogivetheirfeedbackonalltheappearedsubjects.Forthispurpose,Meritsystemhasgivenaninterfacetoeverystudentwithaseparateusernameandpassword.Italsoensureslotofliberty,freenessandconfidentialitytothestudentstoratetheteachingskilloftheirteachersbyconsideringvaried15factors.Itcomputesappraisalresultintheformof30points.Almosteverystaffineachofthedepartmenthasscoredpointsintherangeof20to30whichisaherculeantaskforthemanagementtomakethemintopropergroups.Because,astaffwithapoints24.9belongstoGroup-B,andastaffwithapoints25.0belongstoGroup-Awhichcreateslotofpsychologicalimbalancesamongthestaffmembersifwefollowthistypeoftraditionalgroupingprocedure.Thisistheprimeconcernofthispaper.Instead,offollowingconventionalgroupingmethods,whydon’twefollowfuzzygroupingsothatthestaffswithpoints24.9and25.0belongstosamegroup.Thisisthemainthemeofthispaper.Here,ithasbeenusedthedatasetswhicharefreelyavailablefromourcollegewebapplicationMERITSYSTEM,whichconsistsofnearly330staffmemberswhoarehandlingnearly500+subjectsineachofthesemesters.Oneverysemester,itcollectsthefeedbackandstoresinadatabasewhichcreatesnearly500+records.Clusteringhasbeendonebasedontheserecordstomakeclustersamongstaffmembers.Preprocessingofdataisrequiredatthisstage,because,itisnotpossibletoworkanyofthetoolslikeMat-Lab,R-tooldirectlywiththissoftwareapplication.Itneedsextractionofthisdataandstoresitseparatelyinaseparatedatabase.Forthispurpose,aseparatedatabasewascreatedalongwithafrontendapplicationnamedas―StaffAppraisalSystem‖.ThisapplicationconsistsofmanyinterfacesthroughwhichitacceptsthefeedbackdatamanuallyfromMERITSYSTEM.Later,itgeneratestheappraisalpointsofstaffmembersinexcelformatonwhichwerunmat-labtools.Lastly,therehasbeencomparisonbetweenthemanuallymadegroupswithClusteringofFacultybyEvaluatingtheirAppraisalPerformancebyusingFeedForward35NeuralNetworkApproachCopyright©2017MECSI.J.IntelligentSystemsandApplications,2017,3,34-40groupsobtainedfromneuralnetworktechniques.Itwasfoundthattheaccuracyofthegroupingsobtainedfromneuralnetworkmethodyieldsmoreaccuratethanmanualgroupings.Thisreducespsychologicalimbalancesamongstaffmembersregardingtheirclusters.II.RELATEDWORKTherehavebeenmanyapproachesmadebyexpertsinthefieldofclusteringtechniques.AnEffectiveDataminingusingIncrementallearningNeuralNetworksispresented[15].Here,ithasbeenusedwithasymbolicmethodbyconsideringhoardobjectsandsyntheticobjectsforclusteringprocessapplyingonemployeedataset.Theyfocusedonsomeoftheattributeslikeemployeesalary,commission,age,E-level,zipcode,etc.Even,theyworkedonuniversitydatasetbyincludingmanyrelatedattributeslikeNoofstudents,StudentFacultyratio,Expenses,FinancialAid,No.of.Applications,PercentageofEnrollment,Academicscale,Locationoftheuniversityetc.Here,theresourceallocationnetworkhasbeentrainedwiththesampledataset.Attheendoftrainingresourceallocation,networklearnscomplexdatasetfortherespectivefunction.Thenetworkhasbee