Data Mining Techniques on Course Selection System Application.docx

Data Mining Techniques on Course Selection System Application.docx

《Data Mining Techniques on Course Selection System Application.docx》由会员分享，可在线阅读，更多相关《Data Mining Techniques on Course Selection System Application.docx（10页珍藏版）》请在冰豆网上搜索。

DataMiningTechniquesonCourseSelectionSystemApplication

DataMiningTechniquesonCourseSelectionSystemApplication

[文档副标题]

Abstract:

Thistechniquereportmainlytalksaboutthecourseselectionsystemapplicationbasedondataminingtechniques.Theapplicationcontainssomespecificfunctionsincludingassociationanalysis,classification,clusteringandoutlierdetection.Intheapplication,thesetechniquesareusedbasedontheapplicationusers’requirement,specificationandwhataspecttheyareinterestedin.Moreover,theymainuseroftheapplicationisthepeoplewhoarrangethecourseinformationsuchascourseinformationregistrar（选课管理员）.Ingeneral,thegoaloftheapplicationistoprovidethecourseinformationregistrarwithdetailinformationaboutwhichcourseshouldbeadded,reducedorcanceled,coursepopularityanalysis,operationoncourseandetc.Inthisreport,somebackgroundknowledgewillbeintroducedfirstlyandhowtoapplytheseknowledgeintotheapplicationwillbeexplainedindetail,thatis,detailsofthedesignandimplementationapproacheswillbespecified.Finally,conclusionwillbedrawnattheendofreport.

Keyword:

datamining,courseselection,application

1Introduction

Dataminingisanindispensablestepintheprocessofknowledge-discoveryindatabase.Ingeneral,dataminingisaprocessthatsearchforsomehiddeninformationfrommassivedatabyusingsomespecificalgorithm.Dataminingisusuallyrelatedtosomesubjectssuchascomputerscienceandetc.Byusingstatistics,onlineanalyticalprocessing,informationretrieval,machinelearning,expertsystem（dependonthepastruleofthumb）,patternrecognitionandetc.toachievethesegoalsabove.Byassociatingwithexampleinreallife,thedataminingtechniquescanbeusedinwidelyrange.Inthisexample,courseselectionisacommonsenseincollege.However,differentstudentshavedifferentpreferencesonthecourseselectionwhichleadstosomeprofessorshavemorestudentsthanexpectedorlessstudentthanexpected.Ifthenumberofstudentsismuchmorethanexpected.Theprofessormustapplyforabiggerclassroomoropenanothercourseindifferenttime.Ontheotherhand,ifthenumberstudentsislessthanexpected.Theprofessormustapplyforasmallerclassroomorevencancelthecourse.Inthisway,professorandstudentcanmakethemostofteachingresource.Allthedecisionaremadebycourseinformationregistrarwhocanusethiscourseselectionsystemapplicationtoachievethisgoal.Allinall,theapplicationhelpscourseinformationregistrartoarrangethecourseinamoreefficientway.

2ExistingDatabase

ExistingDatabasemeansthatthecourseinformationregistrarhavetheaccesstothesedata,thatis,thesearerawdataforregistrartousethisapplicationtoanalyze.

✓StudentInformation:

Includingstudentpersonalinformation,studentname,gender,dateofbirth,studentID,classnumber,major,grade,selectionofcourseandetc.

✓ProfessorInformation:

Includingprofessorpersonalinformation,professorname,gender,dateofbirth,professorID,coursetoteachgradeandetc.

✓ClassroomInformation:

Includingclassroominformation,classroomnumber,size,availabletimeandetc.

✓CourseInformation:

Includingcourseinformation,coursename,coursenumber,coursehours,creditandetc.

✓HistoryRecord:

Recordsomedataaboutthecoursethatprofessortaughtincludingstudentlist,professorlistandclasslistofeverycourseinhistorywhicharemassiveofdatatobestoredindatabasetobeanalyzed.

Therelationshipabovecanbedescribeasthegraphasfollow:

3DataMiningTechniques

3.1AssociationAnalysis

3.1.1Theory

Themainaspectofassociationanalysisfrequentpatternanalysiswhichisapattern（asetofitem,subsequences,substructuresandetc.）thatoccursfrequentlyinadataset.Thetheoremisfirstproposedinthecontextoffrequentitemsetsandassociationruleminingwhichaimsatfindinginherentregularitiesindata.Someapplicationssuchasbasketdataanalysis,cross-marking,catalogdesign,salecampaignanalysis,weblog（clickstream）analysis,andDNAsequenceanalysisarerelatedtofrequentpatternanalysis.Moredetail,oneofthescalableminingmethodiscalledApriorialgorithm.Thealgorithmcanbedescribedasaflowchartasfollow:

ByusingtheApriorialgorithm,n-frequentpatterncanbefoundasnchangeswhilenisthenumberofitemsets.Forexample,byanalyzingtheitemboughtofasupermarket,runtheApriorialgorithmandtheclerkwillfindthatpeoplewillbuybeeranddiaperatthesametimewhichhelpthemanagerofsupermarkettoadoptsomesalestrategytopromotethesalevolume.Similarly,theApriorialgorithmalsoapplyforthescenarioofcourseselectionsystemapplication.

3.1.2Implementation

Inthecourseselectionsystemapplication,thecourseinformationregistrarshouldlogintothesystemfirstlyandthenchoosedataminingtechniquesandrawdatabase.Inthissituation,thedataminingtechniquesisassociationanalysisandthedatabaseisstudentandcoursedatabase.Moredetail,theregistrarruntheApriorialgorithmbasedonstudentandcoursedatabaseandsettheitemsetsashewants,forexampletwoitemsets.Then,theapplicationwillshowalotoffrequentpatternsrankedfrommaximumnumbertominimumnumberofpatternswhichhelptheregistrartoanalyzethedataandmakedecision.Forexample,themaximumnumberofpatternsmeansthatthesetwocoursearemostrelatedwhichindicatethatmajorityofstudentswillselectthesetwocourseatthesametime.Moreover,thiswillimpliestheregistrartoarrangethesetwocourseadjacentinthecourseselectionsystemapplicationforstudenttoselectconveniently.Forexample,theregistrarfindsthat60%studentswillselectAdvancedMathematicalandMathematicalModellingatthesametime.Byanalyzingtheresult,theregistrarshouldarrangethesetwocoursesatthesamewebpageforstudenttoselectfromwhichbringsalotofconvenienceforstudenttoselectrelativecourse.

3.2Classification

3.2.1Theory

Theclassificationisalsocalledsupervisedlearningwhichthetrainingdatasuchasobservations,measurementandetc.areaccompaniedbylabelsindicatingtheclassoftheobservations.Moreover,newdataisclassifiedbasedonthetrainingset.Beforeusetheclassificationtheory,categoricalclasslabels（discreteornominal）shouldbepredicted.Moreover,data（constructsamodel）basedonthetrainingsetandthevalues（classlabels）inaclassifyingattributeandusesitinclassifyingnewdatashouldalsobeclassified.Onewaytoclassifythedataisthedecisiontree.Decisiontreeinductionisthelearningofdecisiontreesfromclass-labeledtrainingtuples.Adecisiontreeisaflowchartliketreestructure,whereeachinternalnode（non-leafnode）denotesatestonanattribute,eachbranchrepresentsanoutcomeofthetest,andeachleafnode（orterminalnode）holdsaclasslabel.Thetop-mostnodeinatreeistherootnode.Thebasicalgorithmofdecisiontreeisthegreedyalgorithm.Treeisconstructedinatop-downrecursivedivided-and-conquermanner.Atstart,allthetrainingexamplesareattheroot.Attributesarecategorical（ifcontinuous-valued,theyarediscretizedinadvance）.Examplesarepartitionedrecursivelybasedonselectedattributes.Testattributesareselectedonthebasisofaheuristicorstatisticalmeasure（e.g.informationgain）.Conditionforstoppingpartitioningisasfollow:

✓Allsamplesforagivennodebelongtothesameclass.

✓Therearenoremainingattributesforfurtherpartitioning-majorityvotingisemployedforclassifyingtheleaf.

✓Therearenosamplesleft.

However,howtoselectsuitableattributeisbasedontheinformationgainwhichiscalculatedbasedonthestepsasfollow:

✓LetpibetheprobabilitythatanarbitrarytupleinDbelongstoclassCi,estimatedby|Ci,D|/|D|.

✓Expectedinformation（entropy）neededtoclassifyatupleinD:

✓Informationneeded（afterusingAtosplitDintovpartitions）toclassifyD:

✓InformationgainedbybranchingonattributeA:

Bycalculatingtheinformationgain,selecttheattributewiththehighestinformationgain,nevertheless,suchanapproachminimizestheexpectednumberoftestsneededtoclassifyagiventupleandguaranteesthatasimple（butnotnecessarilythesimplest）treeisfound.Allinall,thedecisiontreeissuitableinthecourseselectioninformation.

3.2.2Implementation

Inthecourseselectionsystemapplication,differentcoursemayhavedifferentstudenttoselectfrom.Byanalyzingthestudentattributeandconstructingadecisiontreerespectively,somepatternscanbefoundwhichishelpfulforanalyzing.Thetrainingdatasetisthestudentdatabase.Firstly,calculatetheinformationgainbasedoneachattributeandselecttheattributewiththehighestinformationgain.Then,runthealgorithmfordecisiontreeandclassifythedata.Moredetails,bycalculatingtheinformationgain,theattributemajorhashighestinformationgain,thatis,themajoristherootofthedecisiontreeandrecalculatetheinformationgain,andfindhighestinformationgainandsoon.Theprocesscanbedescribedasflowchartasflow:

Inthisapplication,byrunningthedecisiontreealgorithm,somepatternscanbefind.Forexample,foreachcourse,suchasjava,thedecisiontreecanbedrawnbasedonthealgorithm.Thedecisiontreeisasfollow:

Inthisdiagram,rootismajorwhichhasthehighestinformationgain,thatis,studentwhosemajoriscomputersciencewillselectthejavacourse.Studentwhosemajorislanguagewilldependongender.Ifgenderismale,thestudentwillselectthec