BAYESIANLINEARREGRESSIONFORPREDICTINGITB ALUMNI’S INCOME(CASESTUDYOFITBTRACER STUDY2021)

• Main Author: Alibasyah Wiriaatmadja, Nur
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/65209
• Institution: Institut Teknologi Bandung

Everyyear,theDirectorateofStudentAffairsofITBconductsaTracerStudy surveyofITBgraduateswiththeaimofprovidingrecommendationstoITBin improvingtheeducationsystemandquality.Thesurveydatacanbeprocessed using apredictivemodelwithmachinelearningapproach,oneofwhichispredicting ITB alumni’ssalariesastheircareerbuildingmilestoneaftercompletingtheir education atITB.ThisFinalProjectaimstobuildaMultipleLinearRegression model topredictthesalaryofITBGraduatesusingthe2021ITBTracerStudydata whose columnshavebeenselected.Thedatausedinthisstudyconsistedof1815 observationsoffourteenpredictorvariablesandsalaryasresponsevariable.To simplify themodelingprocessandtofulfillmodelassumptions,datapreprocessing wasconductedsothatthedatareducedto1803observationsconsistingofeight predictor variablesandoneresponsevariable.Furthermore,theMultipleLinear RegressionmodelwassolvedbyusingBayesianstatisticalapproachtoapproximate the posteriordistributionoftheparametermodel.Metropolis-HastingsAlgorithm are usedasnumericalmethodwhichrequirestheparameterspriordistributionand the likelihooddistributionofthedataasinput.Theposteriordistributionobtained has ameanthatisquiteclosetotheestimatedparametervaluesobtainedusingthe Ordinary LeastSquaremethodasfrequentiststatisticalapproach.Samplingcanbe done fromtheposteriordistributiontomakepredictionsifanewdatasetisknown. Prediction canbedonerepeatedlyusingdifferentsetofparametersobtainedfrom sampling, sothatahistogramofpredictedvaluescanbegeneratedthatismore informativethansinglepredictionusingtheOrdinaryLeastSquaremethod.The proximity ofprediction’smeanfromthetruevaluearevariative.