36 付録2：第98回全米交通運輸調査委員会論文発表会にての発表用ポスター Results and DiscussionResults and DiscussionAcknowledgementsThisstudywassupportedinpartbytheMitsuiSumitomoInsuranceWelfareFoundation2016,JSPSKAKENHIGrantNumberJP17K06612.TheauthorsaregratefultotheTrafficSafetyandCrimePreventionDivision,RegionalPromotionDepartmentofToyotaCitywhoprovidedthecrashdatausedinthisstudy.Meanwhile,theauthorsalsogratitudeAichiPrefecturalPoliceinJapanforcollectingandrecordingthetrafficaccidentdatainToyotaCity.MethodologyBackground and ObjectivesData PreparationThe Transportation Research Board (TRB) 98th Annual Meeting January 13–17, 2019EXAMINING THE ENVIRONMENTAL, VEHICLE AND DRIVER FACTORS ASSOCIATED WITH CROSSING CRASHES OF ELDER DRIVERS USING ASSOCIATION RULES MININGJia Yanga, Keiichi Higuchib, Ryosuke Andoc, Yasuhide Nishihorid/ Email: yang@ttri.or.jp a, c, d Research Department, Toyota Transportation Research Institute, Motoshiro-cho 3-17, Toyota, Aichi 471-0024, Japanb Civil Engineering and Environmental Design Course, Daido University, Hakusui-cho 40, Minami-ku, Nagoya, Aichi 457-8532, JapanPaper Number: 19-00873Research backgroundResearch objectivesModel specificationModel estimationVehicle crash dataThisstudyused5yearsofvehiclecrashrecords(2009-2013)collectedinToyotaCity,Japan.Thetotalnumberofobservationinthecrashdatasetis9,706(from2009to2013)including1,313crashesduetoelderlydrivers(≥65yearsold)and8,393crashedduetonon-elderlydrivers(<65yearsold).ThepositionofvehiclecrashdataisshowninFigure1.NXXS/)()(= (1) NYYS/)()(= (2) NYXYXS/)()(=→ (3) )(/)()(XSYXSYXC=→ (4) )(/)()(YSYXCYXL=→ (5) where, )(XS: support of antecedent X, )(X: number of observation with antecedent X, )(YS: support of consequent Y, )(Y: number of observation with consequent Y, )(YXS→: support of the association rule YX→, )(YX→: number of observation with antecedent X and consequent Y, N: total number of observation in the dataset, )(YXC→: confidence of the association rule YX→ and )(YXL→: lift of the association rule YX→. The lift of rule indicates the frequency of co-occurrence of the antecedent ThedataminingmethodologyonthetransactiondatausingtheassociationrulesminingwasproposedbyAgrawal(Agrawaletal.,1993).Thismethodologyisanassociationdiscoveryapproachusedtodiscovertherelativefrequencyofsetsofitems(i.e.crossingcrashinthisstudy)occurringaloneandtogetherinagivenevent(i.e.acrashobservationinthisstudy).Toimplementthisdataminingtechnology,theApriorialgorithmproposedbyAgrawalandSrikant(1994)isappliedinthisstudy,whichisalevel-wise,breadth-firstalgorithmcountingtransactions.ThefreestatisticalsoftwareRhasapackagecalled“arules”tomakeananalysisofassociationrulesminingusingthisalgorithm.Therefore,thisstudyutilizedapackageof“arules”inopensourcestatisticalsoftwareRtoconducttheassociationanalysis(Hahsleretal.,2005).•Identifythedistinctivecrashpatternduetoelderdrivers,comparedtonon-elderdrivers.•Examinetheenvironmental,vehicleanddriverfactorsassociatedwithcrossingcrashesduetoelderdriversbasedonassociationrulesmining.ThevehiclecrashesduetoelderdrivershavebeenamajorconcernforroadwaytrafficsafetyissueinJapan,sincetheproportionofvehiclecrashesduetoelderdrivershasbeenincreasedupto20%,althoughthenumberofvehiclecrasheshasatrendtodecreasefrom2005to2015.Soreducingthevehiclecrashescausedbyelderdrivershasbecomeaveryimportantissue,especiallyinthefutureagingsociety.Elderdriversaremorelikelytobeinvolvedinvehiclecrashesoccurringinintersectionswithoutsignals,andcrossingcrashestakethelargestproportionamongthecrashtypes.Previousstudieshaveinvestigatedtheassociatedfactorsofcrossingcrashes.However,thesepreviousstudiesarebasedonthetraditionalstatisticalmethodologywhichhasaratherlimitedabilitytorevealtheassociatedrelationbetweenfactorsandcrashes.Asoneofdataminingmethods,associationrulesminingisanidealmethodologytodiscovernewdependencebetweenvariousfactorsandcrashpatternbasedonthetrafficaccidentsdata.Thismethodologywasseldomappliedtoidentifykeyfactorsrelatedtocrossingcrashesofelderdrivers.Urban areaSemimountainous areaFigure 1 Vehicle crashes due to elder and non-elder drivers in Toyota (Digital Road Map) Table 1 Description Statistics of Significant VariablesTable 2 Estimation Results of Association Rules MiningThecrashesofelderandnon-elderdriversdifferencedinlocation,lighting,vehicletype,trafficviolationaswellasthetypeofcrash.Elderdriversaremorelikelytocrashinintersectionswithoutsignalsandinthelightinglevelofdaylight.Theyarealsomorelikelytocausecrashesoflightmotortrucks,maketrafficviolationinwhichtheyfailedtoconfirmsafetyandbeinvolvedincrossingandrightturncrashes.Note:Nooverlapinthe99%confidencelevelindicatesastatisticalsignificance.Forelderdrivers,therewere2ruleshavingthehighestliftvalue(3.446):“Violation=Disobeystopsign→Crossingcrash”,“Violation=Disobeytrafficlights,Location=Intersectionwithsignal,Lighting=Daylight→Crossingcrash”.Thesetworulesindicatedthesinglefactororcombinationoffactorswhichhadthelargestproportionofcrossingcrashinsidecrashtypeforelderdrivers.Fornon-elderdrivers,thehighestliftvalue(lift=4.355)isfoundfora2-productrule:“Violation=Disobeystopsign→Crossingcrash”indicatingthattheproportionofcrossingcrashesinvolvingdisobeystopsighismorethanfourtimesforproportionofcrossingcrashinsidethecrashtype.Comparedtodataminingresultsrelatedtonon-elderdrivers,thedifferentfactorsassociatedwithrelativelylargeproportionofcrossingcrashesincludedlocation(intersectionwithoutsignal),lighting(daylight),roadcondition(dry,other),weather(clear,raining),vehicletype(lightmotortruck)andtrafficviolation(failtoconfirmsafety).Thesedifferentfactorsmightindicatethedifferentcharacteristicsbetweenelderandnon-elderdrivers,andelderdriversmightleadtocrossingcrashesassociatedwithmorefactors,comparedtonon-elderdrivers.Thesefindingscanhelpustomakesomecountermeasurestoimprovethetrafficsafetybyeducatingthem.Aninterestingfindingwasthatthetrafficviolationfailtoconfirmsafetywashighlyassociatedwithcrossingcrashesinsomeoccasions,andtheseoccasionsshouldbesetastheeducationtargetsforelderdriverssinceelderdriverswerelikelytomakeatrafficviolationinvolvingfailtoconfirmsafetyshowninTable1.Thereasonsforthedifferentassociationrulesappliedtoelderdriverandnon-elderonesarelistedasfollowing.Firstly,theproportionofcrossingcrashesconcerningelderdriversarelargerthanthatofnon-elderdrivers.Inthispaper,thethresholdvalueofconfidenceappliedtoassociationrulesminingissetas70%,andtheassociationrulesconcerningnon-elderdriverscannotbeextractedfromthedataset.Secondly,theelderdriverhasalargeproportionoftrafficviolationcalledfailtoconfirmsafety.Here,theviolationcalledfailtoconfirmsafetyishighlyrelatedtothecrossingcrashes,whichisconcludedinthepreviousstudy(Matsuura,2016).ConclusionsResultsofbasicstatisticalanalysisindicatedelderdriversaremorelikelytocrashintheintersectionswithoutsignalsandinthelightingofdaylight.Theyarealsomorelikelytocausecrashesofthelightmotortruck,maketrafficviolationinwhichtheyfailedtoconfirmsafetyandbeinvolvedincrossingandrightturncrashes.Resultsofassociationrulesminingindicatedthattherearemorefactorsassociatedwithcrossingcrashesofelderdrivers,comparedtonon-elderdrivers.Thesefactorsincludecrashlocation(intersectionwithoutsignal),lighting(daylight),roadcondition(dryandother),weathercondition(clearandraining),vehicletype(lightmotortruck)andtrafficviolation(failtoconfirmsafety).Findings from Table 1Findings from Table 2Elder driversNon-elder driversPercentage99% CIsPercentage99% CIsCrashlocationIntersectionwithsignal22.219.2-25.119.017.9-20.2Intersection without signal39.235.7-42.734.132.8-35.4Segment38.635.1-42.146.945.5-48.3Total100.0100.0LightingDaylight61.558.0-64.950.048.6-51.4Dawn24.121.1-27.221.320.2-22.5Night14.411.9-16.928.727.4-29.9Total100.0100.0RoadconditionDry89.587.3-91.787.086.1-88.0Other10.58.3-12.713.012.0-13.9Total100.0100.0WeatherconditionClear77.174.2-80.174.573.3-75.7Raining10.38.1-12.411.810.9-12.7Other12.610.2-14.913.712.7-14.6Total100.0100.0VehicletypeLight motor truck17.114.4-19.74.53.9-5.0Lightmotorcar18.115.3-20.821.920.7-23.1Ordinary motor truck4.63.1-6.110.29.4-11.1Ordinarymotorcar60.256.8-63.763.462.0-64.8Total100.0100.0TrafficviolationInattention22.019.1-25.030.829.5-32.1Fail to confirm safety60.957.5-64.455.153.7-56.5Incorrectoperation5.64.0-7.35.64.9-6.2Failtoobserveobjects2.51.4-3.62.82.3-3.3Disobeytrafficlights3.01.8-4.22.21.8-2.6Disobeystopsign2.31.2-3.31.31.0-1.6Other3.72.3-5.02.21.8-2.6Total100.0100.0CrashtypeHitpedestrian8.16.1-10.05.54.8-6.1Hitfixedobject4.83.3-6.32.92.4-3.3Headon5.23.6-6.83.32.8-3.8Rear end24.621.5-27.744.142.7-45.5Crossing29.025.8-32.223.021.8-24.1Right turn10.48.3-12.67.06.3-7.7Other17.915.2-20.614.313.3-15.3Total100.0100.0AntecedentTypesSupportConfidenceLiftElderdriversViolation=Disobeystopsign2-product0.0231.0003.446Violation=Disobeytrafficlights2-product0.0270.8973.093Location=Intersectionwithsignal3-product0.0270.9723.350Lighting=Daylight4-product0.0231.0003.446Lighting=Daylight3-product0.0230.9383.231Weather=Clear4-product0.0180.9583.303Vehicle=Ordinarymotorcar3-product0.0140.9003.102Weather=Clear4-product0.0110.9333.216Lighting=Daylight&Location=IntersectionwithoutsignalType=Lightmotortruck4-product0.0370.7382.545Violation=Failtoconfirmsafety5-product0.0330.7542.600Road=Dry5-product0.0330.7412.555Road=Other4-product0.0170.7102.446Violation=Failtoconfirmsafety5-product0.0140.7312.518Weather=Raining4-product0.0240.7052.428Violation=Failtoconfirmsafety5-product0.0210.7302.515Non-elderdriversViolation=Disobeystopsign2-product0.0131.0004.355Violation=Disobeytrafficlights2-product0.0200.9043.936Vehicle=Ordinarymotorcar3-product0.0110.9143.982Location=Intersectionwithsignal3-product0.0200.9083.955Note:One6-productrulerelatedtoelderdriverswasnotshowninTable2.
元のページ ../index.html#41