Modling Motorcycl Accidnt on Rural Highway A.K.Sharma, V.S. Landg, and N.V.Dshpand Abstract Motorcyclists ar th most crash-pron road-usr group in many Asian countris including India. Statistics of accidnt on Indian roads rvals that motorcycls accountd for th highst shar in total road accidnts in 20. Earlir studis hav authnticatd th us of Gnralizd Linar modls with Poisson or Ngativ binomial rror structur, for accidnt modling. This study xamins th ffct of road gomtry and traffic variabls on motorcycl crashs using a statistical tchniqu calld as Zro inflatd ngativ binomial rgrssion. Th indpndnt variabls slctd for this study includs accss dnsity (AD), annual avrag daily traffic(aadt), havy vhicl prcntag(hvper), spd variation from modal spd (VARMSP), standard dviation of spd(stdsp), and shouldr width dficincy(swdef). Accidnt pr yar pr km(accr) is takn as dpndnt variabl. Accidnt data collctd btwn 2005-09 ovr a strtch of 00 km of road lngth ar usd for modling. It is obsrvd that shouldr width dficincy, prcntag of havy vhicls in traffic and spd variations hav significant impact on safty of motorcyclist. Kywords Crash rat, accss dnsity, shouldr width dficincy, zro inflatd ngativ binomial I. INTRODUCTION OAD accidnts ar a human tragdy, which involv high R human suffring. Thy impos a hug socio-conomic cost in trms of untimly daths, injuris and loss of potntial incom. Th total numbr of road accidnt in India in 20 was 497,686, out of which 24.4% wr fatal. Th ag profil of accidnt victims rvals that mor than half wr in th wag arning ag group (25-65) []. Motorcyclists ar th most crash-pron road-usr group in many Asian countris including India. Statistics of accidnt on Indian roads rvals that motorcycls accountd for th highst shar in total road accidnts (23.7%) in 20 []. Motorcyclists ar th most vulnrabl road usrs with maximum fatality (26.6%) in total road accidnt. Rural roads wr rportd to hav mor numbr of accidnts (53.5%) than urban aras (46.5%).Also rural roads had mor fatalitis (63.4%) than urban aras (36.6%) and numbr of A.K.Sharma is with Civil Enginring Dpartmnt of Shri Ramdobaba Collg of Enginring and Tchnology, Nagpur(India) Pin-44003, phon- 072-2582844, mobil-09823263.( mail:sharmaakrkn@yahoo.co.in) Dr. V.S.Landg is with Civil Enginring Dpartmnt of VNIT, Nagpur (mail:vslandg@rdiffmail.com) Dr.N.V.Dshpand is with Gurunanak Institut of Tchnology, Nagpur as a Principal. (mail:narndravdshpand@gmail.com) prsons injurd was mor in rural aras (59.4%) as compard to urban aras (40.6%) []. National Highways (on of th rural roads in India, abbrviatd as NH) accountd for 30. % in total road accidnts and 37.% in total numbr of prson killd in 20[]. Prsnt study, on motorizd two-whlrs, is basd on data collctd from National Highway No.6 (NH-6) for a road lngth of 00km btwn Amravati (a city on NH-6) and Lakhni (a township on NH-6). Earlir studis hav authnticatd th us of Gnralizd Linar modls with Poisson or Ngativ binomial rror structur, for accidnt modling. This study prsnts dvlopmnt of prdictiv modls for motorcycl accidnts using Zro Inflatd Ngativ Binomial. % S h a r 30 20 0 0 % shar in total accidnts and prson killd Pdstrian Bicycl Motorcycl Autorickshaw Car,Taxi Truck Buss Fig. Prcntag shar of diffrnt vhicls in total road accidnts and prson killd ( India-20) II. OBJECTIVE Th spcific objctivs of studis ar: 40000 20000 Dvlopmnt of Corrlation btwn road accidnts and gomtric dsign paramtrs of highway along with traffic oprating charactristics for pdstrian accidnt. Evolving nginring rmdial masurs for improving safty on th slctd strtch. Practical rcommndations for improving traffic safty on th said highway. Othr Motor Vhicls Othr Road Usr Typ % shar in total accidnts Prson killd 0 P r s o n K i l l d 33
III. STUDY AREA AND SCOPE Th study ara chosn is National Highway No.6 commonly rfr to as NH-6 or G.E. Road (Grat Eastrn Road),which is a on of th most busy national highways in India. It s a conncting corridor to major stats of India namly Gujarat, Maharashtra, Chhatisgarh, Orissa, Jharkhand and Wst Bngal. NH-6 vntually will b on of th important links of Asian Highway ntwork (AH-46). Th scop of prsnt study is limitd to road sction passing through cntral Indian stats of Maharashtra. Maharashtra, on of th most advancd stats of India, has high traffic accidnt rat. Most of ths accidnts occur on th National highways. IV. METHODOLOGY Mthodology adoptd for th study is as spcifid blow: Idntification of study ara Crash data collction from th law nforcmnt agncy and insuranc companis. Road gomtric and Traffic paramtr data from fild studis Slction of variabls and modling mthods Dvlopmnt of accidnt prdiction modls Tsting of modls, intrprtation of rsults and rmdial masurs suggstions. V. DATA COLLECTION AND ANALYSIS Accidnt data was collctd from th polic stations and insuranc companis. Road gomtry and traffic data was collctd through fild studis and traffic count survy for a road lngth of 00km btwn Amravati City and Nagpur City of Maharashtra Stat in India. For th purpos of collcting road gomtry data, th road was dividd into sgmnts of similar charactristics ranging from 0.2 to 0.6km lngth for curv portion and 0.7 to.6 km for straight portion. Th prliminary analysis of th data ar givn in fig.2 and fig.3. Fig.2 shows a positiv rlation btwn dpndnt variabl and shouldr width dficincy, which suggst that mor dficint shouldrs giv ris to mor accidnts. Fig.3 which givs rlation of havy vhicl prcntag with th dpndnt variabl suggsts that accidnt rats may incras with prcntag of havy vhicls in th traffic stram. Similarly rlationships of othr variabls can b prsntd graphically. VI. VARIABLES USED IN THE MODEL A thorough scrutiny and th prliminary analysis of th data ld to slction of following highly influntial indpndnt paramtrs. Shouldr Width Dficincy: Shouldr provids an ara along th highway for vhicl to stop, particularly during mrgncy. A rport by Zgr t al.[2] indicatd that a pavd shouldr widning of 2 ft pr sid rducs accidnts by 6%. Shouldr width dficincy from a standard minimum (5m in this study inclusiv of both sids) has bn a variabl with significant influnc on saf oprations of traffic and hnc slctd as a variabl. Accss Dnsity: Th availability of accss point is ncssary to commrcial or rsidntial dvlopmnts, usually at th xpns of traffic oprations and th safty of local highway systms. To achiv a good coordination of ths two aspcts, compromiss ar oftn rquird to b mad btwn accssibility and mobility or capacity and safty. On th study ara accss dnsity ranging from 0 to as much as 6 pr km is obsrvd on th said highway. Traffic Volum: Traffic volum is blivd to hav considrabl impact on th crash rat [3]. For this study annual avrag daily traffic (AADT) is usd as a paramtr to indicat traffic volum. Havy Vhicl Prcntag: Thr ar two main traffic rlatd issus associatd with commrcial vhicls, namly: dlays that thy may caus to othr vhicls and safty rlatd impacts [4]. It has bn suggstd by a numbr of authors that th prsnc of a truck in front of any othr vhicl may rsult in th drivr bing mor cautious du to th larg siz of th vhicl and th diminishd sight distancs[5] [6]. Spd Variations: Spot spd is on of th major paramtr that is usd as an indicator of traffic prformanc. Th data collctd shows a wid variation in th spot spd from 25kmph to 60kmph. Th data collctd showd positiv impacts of spd variations on traffic safty. Th spd variations of th vhicls ar incorporatd in trms of variation in spd from modal spd and standard dviation of spd. VII. MODELING OPTION AVAILABLE Various modling tchniqus hav bn trid to modl accidnts aiming for accuracy. Howvr th suitability of th modl dpnds on th data quality and is location spcific. Th modl building mthodology is slctd basd on th availability of data and accuracy of th data. Various modling tchniqus popularly mployd ar as discussd blow : Stochastic Modls : Early in 989; Okamoto t al.[7] suggstd that th occurrnc of traffic crashs follows stochastic distribution. In 990, Garbr t al.[8]dvlopd svral modls to dscrib th occurrnc of crashs using stochastic modling tchniqus, lik Poisson rgrssion(pr) [9] and ngativ binomial rgrssion(nbr)[0][].mor stochastic modls wr also proposd othr than Poisson rgrssion modls and Ngativ Binomial Rgrssion Modl, which includd Zro Inflatd Poisson Rgrssion Modl and Zro Inflatd Ngativ Binomial Rgrssion Modls. This papr prsnts modls build using Zro Inflatd Ngativ binomial Rgrssion Zro Inflatd Ngativ Binomial 34
Rgrssion Modls: Transportation safty analysts hav typically justifid th us of Zro Inflatd (ZI) modls bcaus of th improvd statistical fit compard to traditional Poisson and NB modls. Zro Inflatd rgrssion modls ar two rgim modls. First probability modl govrns whthr a count numbr is zro or positiv numbr, calld as inflat modl. Thn th positiv part of th distribution is dscribd by suitabl stochastic distribution, calld as bas modl. In Miaou s[2] study, th ngativ binomial rgrssion modl was of th form; y i = 0,,2 wwwh ppppppppppp Γ α + y i Γ ( ) α λ i α ( ) y i () α Γ(y i + ) λ i = β ix i wh x i ii i th covvvvvvv aaa β i ii th rrrrrrrrrr ccccccccccc For Zro Inflatd Ngativ Binomial rgrssion y i = 0 wwwh ppppppppppp p 0 + ( p 0 )( ) α (2) y i =,2 wwwh ppppppppppp Γ ( p 0 ) α + y i Γ ( ) α λ i α ( ) y i (3) α Γ(y i + ) Whr p 0 can b rprsntd by probability modl incorporating th ffcts of covariats, such as logit modl. r w i p 0 = (4) + r w i r is th cofficint matrix and w i is th i th covariat.γ(.) is Gamma function; and α is th rat of ovr disprsion. VIII. MODEL SELECTION CRITERIA Maximum liklihood stimation mthod has bn mployd widly in stimating Poisson, ngativ binomial [3] and zro inflatd rgrssion modls. Akaik Information Critrion (AIC) [4] and Baysian Information Critria(BIC) was usd to judg th prformanc of th modl. Smallr th AIC and BIC valus, th bttr is th modl. AIC= -2Log L +2K.. (5) Whr Log L is th log liklihood; K is th numbr of stimatd paramtrs. IX. VARIABLE SELECTION Various combinations of th variabls slctd wr trid th bst combinations ar givn in Tabl I. Accidnt rat pr yar pr km was chosn as dpndnt variabl. Crash rat is dfind as CR i = TA i / L i / NY, whr CR i = Crash Rat on sgmnt, TA i = Total accidnts on sgmnt i, L i = Lngth in Km. of sgmnt i, NY = Numbr of Yars X. RESULTS Th rsults obtaind aftr analysis of data, using SPSS softwar, ar shown in TABLE II and TABLE III. In ordr to s th prformanc of th modl th cofficint of variabls has to b xamind. Modls with logical algbraic signs of th variabls wr slctd. Along with th strong statistical tools, propr nginring judgmnts ar rquird to dcid upon th slction of final modl. For xampl, a positiv sign of shouldr width dficincy suggsts that mor shouldr width dficincy will b rsponsibl for mor accidnts and it is logically accptabl. A. Bas Modl Acc/Yr/Km= 2.2 + 0.33 SSSSS + 0.079 HHHHH +0.044 VVVVVV + 0.0000288 AAAA (6) λ = 2.2+0.33 SWDEF+0.079 HVPER+0.044 VARMSP+0.0000288 AADT (7) Substituting avrag valus for variabls, w gt λ = 0.562 B. Inflat Modl p 0 = + 3.52+.688 SWDEF+0.4 HVPER (8) 3.52+.688 SWDEF+0.4 HVPER Substituting avrag valus for variabls, w gt p 0 = 0.57 C. Final Modls Modl to prdict frquncy of zro accidnts P(y = 0) = ε [ p o + ( p 0 ) ) α (9) Modl to prdict frquncy of accidnts othr than zro Γ P(y) = ε ( p 0 ) α + y i Γ α Γ(y i + ) α α λ i (0) Whr ε is xposur trm = total no. of road sctions(28 in prsnt study) α is ovrdisprsion paramtr =0.62 (for prsnt data collctd ) TABLE I VARIABLE IN THE MODEL Modl Indpndnt Variabls No. SWDEF, HVPER, VARMSP, AADT 2 SWDEF, HVPER, VARMSP 3 SWDEF, HVPER, STDSP 4 SWDEF,HVPER,VARMSP,AD Dpndnt Variabl y i Accidnts/Yr/Km (Crash Rat) 35
Fig. 2 Variation of accidnt rat with shouldr width dficincy Fig. 3 Variation of accidnt rat with prcntag of havy vhicls TABLE II PARAMETER ESTIMATES FOR BASE MODEL (USING SPSS) Variabl Modl Modl 2 Modl 3 Modl 4 Intrcpt -2.2-2.028-2.64-2.025 SWDEF 0.33 0.42 0.46 0.2 HVPER 0.079 0.082 0.089 0.082 VARMSP 0.044 0.04 0.040 AD ------ ------ ------- 0.043 AADT 0.0000288 ------ ------- ------- STDSP ----- ------ 0.035 ------- AIC 6.52 66.49 77.55 65.46 BIC 77.93 79.26 89.32 8.88 TABLE III PARAMETER ESTIMATE FOR INFLATE MODEL Inflat Variabl SWDEF HVPER Intrcpt Cofficint +.688 +0.4-3.52 Accidnt TABLE IV PREDICTED AND OBSERVED FREQUENCY Prdictd Obsrvd 0 89.8 7 4.06 24 2 9.55 4 3 5.99 2 4 3.6 5 2. 6.2 7 0.69 0 8 0.38 0 9 0.2 0 0 0.2 0 R-Squar 0.959 Fig.4 Obsrvd vs prdictd frquncy 36
XI. CONCLUSION Zro inflatd ngativ binomial rgrssion was proposd to stablish an mpirical rlationship btwn motorcycl accidnts and highway gomtric and traffic paramtrs. Th rsult of this study can vntually b mployd to idntify th locations with crtain numbrs of accidnt frquncis undr diffrnt gomtric and traffic conditions for national highway numbr-6 (NH-6) in India. Th rsults obtaind hr provid valuabl insight into th undrlying rlationship btwn risk factors and vhicl accidnts. Th modl slctd may b usd to dvlop stratgis for nhancing safty with optimum us of rsourcs. Th ffcts of various variabls can also b quantifid from th paramtr stimats of th modls. Motorcycl accidnt is gratly influncd by shouldr width dficincy on a particular road sction. Rducing th shouldr width dficincy by m may rduc th accidnts by about 24%. Evry 2% incras in havy vhicl traffic may incras th accidnts by about 28%. Similarly spd variation is also a influntial factor affcting accidnts. XII. SCOPE FOR FUTURE WORK Th rgrssion mthod may b usd to modl traffic accidnt for diffrnt st of crash data collctd from diffrnt sctions of th sam or som othr highway. Th mthod may b compard with othr mthods lik ANN. Futur work might also focus on improving th prdiction prformanc of th ZINB modls. [3] Landg,V.S.,Jain S.S. & Parida,.(2006), Modling traffic accidnts on two lan rural highways undr mixd traffic conditions, 87th Annual Mting of Transportation Rsarch Board,(CD- Rom). [4] Sharma A.K., Landg V.S.(202), Pdstrian Accidnt Prdiction Modl For Rural Road, Intrnational Journal of Scinc and Advancd Tchnology Volum 2, No 8. A.K.Sharma,born on 28-2-64,graduatd in civil Enginring from NIT Raipur(Chattisgarh,India) in 987. H compltd his post-graduat in Highway and Traffic Enginring from IIT Kharagpur (India) in 989. Prsntly working at Shri Ramdobaba Collg of Enginring and Managmnt, Nagpur, as Associat profssor in Civil Enginring Dpartmnt. His main fild of work has bn pavmnt and traffic nginring. H has guidd many undrgradut and post graduat dissrtations. Prsntly h is prsuing his Doctoral rsarch in th fild of traffic safty. V.S.Landg, born on 02-0-68, graduatd from RKNEC, Nagpur (India) in 99.H compltd his post graduat studis in Transportation Enginring from BITS Pillani(Rajasthan, India) in 993. H compltd his Doctoral dgr from IIT Roork(India) in 2006. Prsntly working at VNIT, Nagpur as Associat profssor in Civil Enginring Dpartmnt. His main fild of work has bn traffic nginring particularly in th fild of traffic Safty. H is a mmbr of th Stat Tchnical Agncy for Pradhan Mantri Gram Sadak Yojna for Vidarbha Rgion in India H has guidd many undrgraduat and post graduat dissrtations. N.V.Dshpand, born on 8-07-956, graduatd from VNIT, Nagpur (India).H compltd his post graduat studis in Structural Enginring from VJTI, Mumbai (Maharashtra, India). H compltd hisdoctoral d gr from VNIT Nagpur (India). Prsntly working at GNIET Nagpur as Profssor and Principal. His main fild of work has bn Structural Engin ring. H has guidd many undrgraduat and post graduat and Doctoral Thsis. REFERENCES [] Road accidnt in India (20) (Ministry of Road Transport and Highway). [2] Zgr, C. V., Dn, R. C. & Mays, J. G. (98) Effct of lan and shouldr widths on accidnt rduction on rural two-lan roads. TRR 806.pp 33-43. [3] Karlaftis M.G. & Golias, I.(2002), Effct of road gomtry and traffic volums on rural roadway accidnt rats, Accidnt Analysis and Prvntion vol.34, Issu 3, pp. 357-365. [4] Kramms, R. and Crowly, K. Passngr Car Equivalnts for Trucks on Lvl Frway Sgmnts. Transportation Rsarch Rcord, 09, pp. 0-7, 987. [5] Molina, C. Dvlopmnt of Passngr Car Equivalncis for Larg Trucks at Signalisd Intrsctions. ITE Journal, Novmbr, 987. [6] Kocklman, K. and Shabih, R. Effct of Light-Duty Trucks on th Capacity of Signalisd Intrsctions. Journal of Transportation Enginring, 26(6),pp. 506-52, 2000. [7] Okamoto (989), A Mthod to cop up with random rrors of obsrvd accidnt rats Safty litratur, Vol.4, pag 37-332. [8] Garbr, N.J., Wu, L(989), Stochastic modls rlating crash probabilitis with gomtric and corrsponding traffic charactristics data, Rsarch rport No UVACTS-5-5- 74 Cntr for transportation studis at th Univrsity of Virginia. [9] Jovanis, P. & Chang, H(986)., Modling th rlationship of accidnts to mils travld, Transportation Rsarch Rcord 068, pp 42-5. [0] Poach, M. & Mannring, F.(996), Ngativ binomial analysis of intrsction accidnt frquncis ASCE s Journal of Transportation Enginring Vol.22 No. 2, pp. 05-3. [] Ziad Sawalha & Tark Sayd (2003) Statistical Issus In Traffic Accidnt Modling TRB annual mting (CD-ROM). [2] Miaou, S.P.(994) Th rlationship btwn truck accidnts and gomtric dsign of road sctions: Poisson vrsus ngativ binomial rgrssions, Accidnt Analysis and Prvntion vol. 26, issu 4, pp. 47-482 37