基于人工神经网络和统计模型的预测(IJEME-V6-N3-3)

I.J.EducationandManagementEngineering,2016,3,20-32PublishedOnlineMay2016inMECS()DOI:10.5815/ijeme.2016.03.03Availableonlineat(wideruseintheinflationmeasurement),andasameansofregulatingincome.ItisalsousedasasupplementforstatisticalchainstopredictfuturevalueindicesinordertomakesurethatthedataaccuratelyreflectthepatternspurchasedbytheYemeniconsumer.Inthispaper,weproposeamodifiedartificialneuralnetworkmethodtopredicttheindicesofconsumerintheRepublicofYementothepricesoftheperiodfrom01/01/2005till01/01/2014.Theresultsofusingtheproposedmethodiscomparedtoaclassicalstatisticalmethod.Theproposedmethodisbasedonartificialneuralnetworks,namely,backpropagationwithadaptiveslopeandmomentumparametertoupdateweights.However,thestatisticalmethodisBox-Jenkinsmodelwhichisusedtopredicttimeseries.Theexperimentalresultsshowthatartificialneuralnetworksgivesbetterpredictivevaluesduetotheirabilitytodealwiththenonlinearandstochasticdatabetterthantraditionalstatisticalmodelingtechniques.IndexTerms:Forecasting,Timeseriesmodels,Neuralnetworks,Box-Jenkins,Consumerpriceindex,Backpropagation,Adaptiveslope,Momentumparameter.©2016PublishedbyMECSPublisher.Selectionand/orpeerreviewunderresponsibilityoftheResearchAssociationofModernEducationandComputerScience.1.IntroductionPredictionoftime-seriesisoneofthecriticalareaswherethereismuchuseofapplicationsofArtificialNeuralNetworks(ANNs)asanalternativeorequalmethodtothetraditionalstatisticalmethodswhichareusedinthepredictionoftime-seriessuchasmoving-average,exponentialsmoothingBox-Jenkinsmodels[1].Thesetraditionalmethodsareknowningeneralastimeseriesanalysismethods.Theneuralnetworksmodelshavecompetedthetraditionalforecastingmethodsusedinthepredictionandevengavebetteraccuracyresultsin*Correspondingauthor.E-mailaddress:ForecastingusingArtificialNeuralNetworkandStatisticsModels21manycases[1,2].Researchershavebeeninterestedinthedevelopmentofvariousmethodsforthepurposeofforecastingusingneuralnetworksasoneofthenewestusedmethods.Itisstillongoingresearchinthisareatoinvestigatetheeffectivenessofthismethod.Mostneuralnetworks;whichallowlearningfromexperienceandpastexperimentstoinfernewones,areusedastoolstoanalyzethedataforthesameareascoveredbytraditionalstatisticalmethods.Neuralnetworksgiveasuitablewaytorepresentrelationshipsbetweenvariableswhicharedifferentfromthetraditionalmethodsandconsideredasmodernstatisticaltools.Theforecastingprocessanalyzesthedatapriortothephenomenonbeingstudiedtoidentifythegeneralpatternofthisphenomenoninthefuture.Thisisoneofthebasicoperationsoftheneuralnetworks,i.e.patternsidentificationandanalysis.Neuralnetworksarenon-linearflexiblefunctionsthatdonotrequiretheavailabilityofrestrictiveassumptionsabouttherelationshipbetweenthedependentandtheindependentvariables.Inadditiontotheiraccuracywhenusedonparametricdata,neuralnetworkscanbeusedfornonparametricorsmall-sizeddata.Whatdistinguishesnetworksmodelsisthelackofanyassumptionsorpreconditionswhenappliedinthefieldofforecastingasinthestatisticalmethodsthatsomeassumptionsshouldbefulfilledbeforeapplyingthem.Forexample,inBoxJenkinsmodelstheassumptionofstabilitymustbefulfilledbeforeconstructingthemodel.Thisassumptionisnotrequiredduringconstructingthenetworksmodels.Therearemanysuccessfulapplicationsofneuralnetworksinvariousareas,suchasmedicine,engineering,banks,insuranceandotherbusinesspractices,andtheyareconsideredasimportanttoolstoinvestorsforthepredictionofinvestmentbehaviorandchoosingthebestinvestmentalternatives.Theuseofneuralnetworkstopredictthetime-seriesbeganattheendoftheeightiesandthefirstattemptwasin[3,4,5],whousedtheperceptronmultilayerandthebackpropagationalgorithminthepredictionofunstabletimeseries.In[6],theauthorsintroducedastudythatsupportstheuseofneuralnetworksinthepredictionoftime-seriesandexplainstheuseofthebackpropagationalgorithmtotrainthenetwork.ThisstudyhasgivenbetterresultswhencomparedtomanyofthetraditionalstatisticalmethodsasRegressiveLinearorBox-Jenkins.ConsumerPriceIndex(CPI)isdefinedastheaverageofthechangeableconsumerpricesfortheirdailyliferequirements.Thesepricesarecollectedeverymonthtogetequaltimeperiodstocalculatetherateofchangeinthesepricesincrease.Theinflationamountiscalculatedusingstatisticalmethods,TheresultsofthesedataaffecttheeconomicchangeintheconsumerandsocietyintheRepublicofYemen[7].predictingtheprices