Innovation
Advanced machine learning solutions for diverse predictive applications and insights.
Inrecentyears,theIntelligenttransportationssystem(ITS)hasreceivedconsiderable
attention,duetohigherdemandsforroadsafetyandefficiencyinhighlyinterconnected
roadnetworks.AsanessentialpartofITS,trafficpredictioncanprovidesupportin
manyaspects,suchasroadrouting,trafficcongestioncontrol,etc.Toprovideamore
comprehensiveoverviewoftheroleoftrafficforecastinginITSsystems,wewillfirst
introducethecorrespondingITSapplicationsanddiscusshowtrafficforecastingcan
improvetheperformanceoftheseapplications.Next,wewillintroducethegeneral
predictionprocedureaswellassomebasicconceptsoftrafficflowprediction,followed
byadescriptionofageneralframeworkforimplementingthetrafficflowprediction.
Inthissurvey,mainlytwosortsofpredictionmethodsarefocused,statistics-based
andmachinelearning(ML)-based..ThesetwotypesofapproachesaremoreusedinITS
trafficflowpredictionstheseyears,andservicefordifferentcontexts.Generally
speaking,thestatistics-basedmodelshavebettermodelinterpretability,butthe
rigorousmodelstructurelimitstheadaptability,whileML-basedmodelsaremore
flexible.
Asmentionedpreviously,ML-basedmodels,thathaveadeepconnectionwithvarious
fields,suchaspatternrecognition,statisticallearning,datamining,computervision,
speechrecognition,andnaturallanguageprocessing,etc.,havecaughtconsiderable
attentioninrecentyears.Multi-disciplinednaturemakesthesemethodshavebetter
adaptabilityfordifferentcontexts.Accordingly,howtodesignapracticalML-based
approachforsupportingtheITShasbecomeahotresearchtopic.InOtherprediction
algorithmsApartfromthestatistics-basedmodelsandML-basedapproachesdiscussed
above,someotherhelpfulalgorithms,suchasKalmanfilterandHiddenMarkovchain,
alsobenefittheshort-termtrafficflowprediction.Thesemethodsareuseful,butfew
mentionedinpaststudies,soitisvitaltoputthemtogethertointroduceinthis
section.

