Journal of Safety Research

JSR-01556; No of Pages 15
Journal of Safety Research xxx (xxxx) xxx
Contents lists available at ScienceDirect
Journal of Safety Research
journal homepage: www.elsevier.com/locate/jsr
1Q2 Relationship between road traffic features and accidents: An application

2 of two-stage decision making approach for transportation engineers
F
Q43Q3 Syyed Adnan Raheel Shah, a,b,⁎ Naveed Ahmad, a Yongjun Shen, c Mumtaz Ahmed Kamal, a Muhammad Aamir
O
4 Basheer, b Tom Brijs b
5 a
Taxila Institute of Transportation Engineering, Department of Civil Engineering, University of Engineering & Technology, Taxila 47050, Pakistan
6 b
O
Transportation Research Institute (IMOB), Hasselt University, Agoralaan, B-3590 Diepenbeek, Belgium
7 c
School of Transportation, Southeast University, Sipailou 2, 210096 Nanjing, China
8
R
9 a r t i c l e i n f o a b s t r a c t
P
10 Article history: Introduction: An efficient decision-making process is one of the major necessities of road safety performance 20
11 Received 21 February 2018 analysis for human safety and budget allocation procedure. Method: During the road safety analysis procedure, 21
12 Received in revised form 28 September 2018 data envelopment analysis (DEA) supports policymakers in differentiating between risky and safe segments of 22
13 Accepted 17 January 2019
a homogeneous highway. Cross risk, an extension of the DEA models, provides more information about risky seg- 23
14 Available online xxxx
D
ments for ranking purpose. After identification of risky segments, the next goal is to identify those factors, which 24
19
18
17
16
15
32 Keywords:
are major contributors in making that segment risky. Results: This research proposes a methodology to analyze 25
E
33 Transportation road safety performance by using a combination of DEA with the decision tree (DT) technique. The proposed 26
34 Roads methodology not only provides a facility to identify problematic road segments with the help of DEA, but also 27
35 Decision making identifies contributing factors with the help of DT. Practical applications: The applicability of the proposed 28
T
36 Accidents model will help policymakers to identify the major factors contributing to road accidents and analysis of safety 29
37 Risk evaluation performance of road infrastructure to allocate the budget during the decision-making process. 30
38
Q6
C
DEA-DT © 2019 National Safety Council and Elsevier Ltd. All rights reserved. 31
42
40
39
E
41
43 1. Introduction 2014” (CARE, 2016). Road safety performance of Belgium shows that 60
R
it is in top 10 worst countries among the peer European countries, as 61

44 Within the European region, road accidents on motorways are a shown in Fig. 1. 62
R
45 major contributing factor in road fatalities; this study conducted a safety However, a realistic approach tells the story that it should be nearer 63
46 performance analysis for European motorways to discover the causes of to the European average and the target should be, at least, the safest 64
47 these accidents. In addition to safety, traffic stream characteristics and country in the Schengen region. Road safety administration profes- 65
O
48 road infrastructure geometric design feature are the major focus of an- sionals have constantly tried to depict an unequivocal point of view in 66
49 alyzing core-contributing factors of increasing road accidents. World- both quantitative and subjective road safety assessment criteria by set- 67
C
50 wide, an estimated 1.2 million people are killed in road crashes each ting reliable systems. 68
51 year and as many as 50 million are injured (WHO, 2004); it is necessary Usually, this type of analysis is dependent on the stage-wise safety 69
52 to improve the standard of road traffic conditions and to improve the performance of a country at almost four levels: federal, regional, provi- 70
N
53 road safety conditions for the betterment of human life. In this battle, sional, and municipal level. At the federal level, decision making is re- 71
54 the European Commission fixed the aspiring goal of halving the number lated to motorways, and overall performance of Belgium among 72
U
55 of road fatalities by 2010 in its White Paper “European transport policy European countries is not satisfactory on motorways. Previously, 73
56 for 2010: time to decide” of 2001. A new target for 2020 to halve the the focus of road safety investigation research was on the local level 74
57 number of road deaths compared to 2010 was set by the EU in its in Europe, including Belgium (Eksler & Lassarre, 2008; Mohan, 75
58 “Road Safety Programme 2011–2020.” It is estimated that the number Bangdiwala, & Villaveces, 2017). “Almost 26,000 people were killed in 76
59 of road accident fatalities in the EU fell by 42% between 2005 and road accidents on motorways in the European Union countries between 77
2005 and 2014. This number corresponds to 7% of all road fatalities in 78
those countries. There were 3,558 road accident fatalities on motorways 79
⁎ Corresponding author at: Transportation Research Institute (IMOB), Hasselt in 2005, and the number fell by 48% in 2014 (1,865).The total number of 80
Q5 University, Agoralaan, B-3590 Diepenbeek, Belgium.
road accident fatalities in the European Union countries also fell signifi- 81
E-mail addresses: syyed.adnanraheelshah@uhasselt.be (S.A.R. Shah),
n.ahmad@uettaxila.edu.pk (N. Ahmad), shenyongjun@seu.edu.cn (Y. Shen), cantly over the same decade, by 42%. Although the overall number of 82
dr.kamal@uettaxila.edu.pk (M.A. Kamal), tom.brijs@uhasselt.be (T. Brijs). road fatalities decreased rather steadily, the trend for motorway fatalities 83
https://doi.org/10.1016/j.jsr.2019.01.001
0022-4375/© 2019 National Safety Council and Elsevier Ltd. All rights reserved.
Please cite this article as: S.A.R. Shah, N. Ahmad, Y. Shen, et al., Relationship between road traffic features and accidents: An application of two-
stage decision makin..., Journal of Safety Research, https://doi.org/10.1016/j.jsr.2019.01.001
2 S.A.R. Shah et al. / Journal of Safety Research xxx (xxxx) xxx
F
O
Fig. 1. Percentage of road crash fatalities of all fatalities by country (2014) (CARE, 2016).
O
84 has been more variable. The most significant reduction of the number of Yeo, Jang, Skabardonis, & Kang, 2013; Zhou & Sisiopiku, 1997), speed 109
85 110
R
fatalities on motorways occurred between 2007–2008. Spain had the (Aljanahi, Rhodes, & Metcalfe, 1999; Elvik, 1997; Elvik, Christensen, &
86 highest percentage of fatalities on motorways in 2014 in the EU (17%), Amundsen, 2004; Garber & Ehrhart, 2000; Golob & Recker, 2004; 111
87 followed by Belgium (15%), Slovenia (13%), and the Netherlands (12%). Hamzeie, Savolainen, & Gates, 2017; Imprialou et al., 2016; Kononov, 112
P
88 By contrast, the lowest proportion of fatalities occurring on motorways Lyon, & Allery, 2011; Malyshkina & Mannering, 2008; Pande & Abdel- 113
89 was in Romania (1%) and Poland (2%)” (EU, 2016). According to the Aty, 2006), flow (Aljanahi et al., 1999; Garber & Ehrhart, 2000; Golob 114
90 European Road Safety Observatory, in 2014 fatality rate was higher in et al., 2004; Golob et al., 2008; Golob & Recker, 2003; Kononov et al.,
D 115
91 Belgium (9.6) than in the other European countries and hence higher 2011), and geometric design (Fu, Guo, Yuan, Feng, & Ma, n.d.; Garber & 116
92 than the average rate (3.7) of the EU countries (EU, 2016) as shown in Ehrhart, 2000; Karlaftis & Golias, 2002; Milton & Mannering, 1996; 117
93 Fig. 2. Therefore, it is necessary to focus on motorways of Belgium. Noland & Oh, 2004; Wang, Hallmark, Savolainen, & Dong, 2017). For mo- 118
E
94 As a traffic engineer and decision maker during the safety analysis torways, during road safety analysis procedure it is necessary to prioritize 119
95 procedure, contributing factors like geometric design, speed control, and the problematic segments of a road and then to the identify contributing 120
T
96 traffic flow characteristics are of major concern. In previous studies, rela- factors as well for an econometric-based decision-making. 121
97 tionship of road crashes fatalities (Dadashova, Arenas-Ramírez, Mira- In this paper, we propose a two-stage DEA-DT approach to analyze 122
C
98 McWilliams, & Aparicio-Izquierdo, 2016; Givehchi, Hemmativaghef, & the road safety conditions of motorways. The proposed methodology 123
99 Hoveidi, 2017; Golob, Recker, & Alvarez, 2004; Golob, Recker, & Pavlis, will be applied to overcome the safety problem for motorways. An illus- 124
100 2008; Imprialou, Quddus, Pitfield, & Lord, 2016; Kaye, Lewis, & Freeman, trative application of the methodology to two motorways of Limburg 125
E
101 2018; Lord, Manar, & Vizioli, 2005; Luoma & Sivak, 2007; Ogwueleka, province in Belgium is given as a case study. 126
102 Misra, Ogwueleka, & Fernandez-Sanz, 2014; Pande & Abdel-Aty, 2006;
R
103 Wang, Quddus, & Ison, 2009; Yasin Çodur & Tortum, 2015) have been 2. Literature review 127
104 studied in context of volume/capacity (Golob & Recker, 2004; Lord,
105 Manar, & Vizioli, 2005; Pande & Abdel-Aty, 2006; Zhou & Sisiopiku, 2.1. Road accidents and safety management 128
R
106 1997), vehicles miles traveled (Abdel-Aty, Lee, Siddiqui, & Choi, 2013;
107 Dadashova et al., 2016; Frantzeskakis & Iordanis, 1987; Jovanis & Chang, Road accidents standout as a threatening health problem. Without 129
O
108 1986; Wang & Kockelman, 2013), vehicles hours traveled (Martin, 2002; new or enhanced improvements, it is anticipated that with the 130
C
N
U
Fig. 2. Fatalities rates per million population by country for Motorways in Europe (2005 & 2014) (EU, 2016).
S.A.R. Shah et al. / Journal of Safety Research xxx (xxxx) xxx 3
131 increased use of vehicles, it will be the fifth leading cause of death by the et al., 2008; Golob & Recker, 2003; Kononov et al., 2011), and geometric 192
132 end of 2030 (OECD, 2002; WHO, 2004; WHO, 2006; WHO, 2015). design (tangent, horizontal, vertical curves) (Fu et al., 2011; Garber & 193
133 Worldwide, the statistics on road accident deaths is 1.2 million, along Ehrhart, 2000; Karlaftis & Golias, 2002; Milton & Mannering, 1996; 194
134 with 20–50 million injuries per year. One of the leading indicators Noland & Oh, 2004). In this investigation, identification of high risk seg- 195
135 used for road safety performance measurement is fatality rate (fatalities ments for traffic crashes (i.e., distinguishing areas with high constant 196
136 per number of population or number of registered vehicles) (Al Haji, crash potential) has been the focus. Different data sources were adopted 197
137 2005; Wegman et al., 2008). In last decade, due to lack of information during traffic data and safety analysis such as provisional road adminis- 198
138 about road safety performance indicators, a composite road safety trations (Wegman, 2014). Applying circle information to anticipate 199
139 index was developed (Al Haji, 2005; ETSC, 2001; Wegman et al., crashes is ongoing and in preparatory stages. In addition, the idea of 200
140 2008). Road safety and its management is a complex system that is crash antecedents guessed that the probability of a crash is fundamen- 201
141 governed in five major dimensions (Haddon Jr, 1980; Wang & Chen, tally influenced by the transient turbulence of movement stream (Lee, 202
142 2012a; Wang & Chen, 2012b). These five dimensions are not completely Saccomanno, & Hellinga, 2002). They thought of elements like speed 203
143 autonomous of each other, and each dimension is impacted by numer- variety along the length of the roadway (i.e., distinction between the 204
144 ous components and indicators (Al Haji, 2005; Rumar, 1999; Wang & speeds upstream and downstream of the crash area), and furthermore 205
F
145 Chen, 2012a; Wang & Chen, 2012b). Choosing an arrangement of safety over the three paths at the crash area (Lee et al., 2002). 206
146 performance indicators, which should fill in as a strong tool of
O
147 benchmarking for policymakers (Wegman et al., 2008), is an unpredict- 2.3. Road accident data analysis models 207
148 able issue. The decision of every indicator is pivotal and primarily relies
149 upon the sort, accessibility, and nature of information/data being gath- Previously, different statistical techniques were used for accident data 208
O
150 ered (Papadimitriou, Yannis, Bijleveld, & Cardoso, 2013). In addition, analysis. Those techniques were Multiple Linear Regression (Mustakim & 209
151 the chosen indicator should be able to fit to join in a composite index, Busu, 2007; Mustakim & Fujita, 2011), Poisson regression (Joshua & 210
R
152 which can consolidate all the significant parameters in a compact and Garber, 1990; Lord, Washington, & Ivan, 2005; Wang & Kockelman, 211
153 comprehensive way (Wegman et al., 2008). Moreover, the composite 2013), Binary Logistic Regression (Al-Ghamdi, 2002; Dissanayake & Lu, 212
154 index should be designed to be adjusted as acceptable and balanced 2002; Kim, Lee, Washington, & Choi, 2007; Sze & Wong, 2007; Wang, Li, 213
P
155 (Al Haji, 2005). Researchers applied the method of risk calculation hav- Chen, & Ni, 2011), Multinomial Regression (Shankar & Mannering, 214
156 ing basic concept as shown in Eq. 1 (Shah et al., 2018). 1996; Ye & Lord, 2011), Tobit (Anastasopoulos, 2016; Anastasopoulos, 215
Mannering, Shankar, & Haddock, 2012; Anastasopoulos, Shankar,
D 216
Road Safety Outcome Haddock, & Mannering, 2012; Anastasopoulos, Tarko, & Mannering, 217
Risk ¼ ð1Þ
Exposure 2008), negative binomial (NB) regression (Chang, 2005; Chang & Chen, 218
158 2005; Ladron de Guevara, Washington, & Oh, 2004; Lord & Mannering, 219
E
With the increase of thoughtful indicators, number of indicators 2010), probit models (Kockelman & Kweon, 2002; Ye & Lord, 2011), Sen- 220
159 have increased to produce a composite road safety index (Al Haji, sitivity Analysis (Geurts, Wets, Brijs, & Vanhoof, 2004; Wang & Chen, 221
T
160 2005; ETSC, 2001; Hakkert, Gitelman, & Vis, 2007; Wegman et al., 2012b), Analysis of Variance, Zero-inflated Poisson (Aguero-Valverde, 222
161 2008). Researchers have explained the criteria of building a complete 2013; Anastasopoulos, 2016; Dong, Clarke, Yan, Khattak, & Huang, 2014; 223
C
162 composite road safety index to replace the conventions methods of Lord, Washington, & Ivan, 2005), and negative binomial (Chang, 2005; 224
163 measuring road safety performance (Wegman et al., 2008). Chang & Chen, 2005; Dong et al., 2014; Ladron de Guevara et al., 2004). 225
These models were effective but the problem was with multiple inputs 226
E
164 2.2. Relationship between safety and road traffic features and multiple outputs (i.e., in case of multiple dependent and independent 227
variables it was difficult to rank and compare the performance of decision 228
R
165 The common way to deal with traffic safety investigation (Abdel-Aty making units [DMUs]). For example, during the analysis of a highway, it 229
166 & Pande, 2007) has been to set up connections between the activity was difficult to select the worst and best segment in the presence 230
167 attributes (e.g., flow, speed), roadway, and natural conditions (e.g., ge- of a large data set. However, after advancement in safety research, 231
R
168 ometry of the expressway), and crash event (Abdel-Aty & Pande, during the European Safety Net Project (2005) (SafetyNet, 2005) on 232
169 2005). The inadequacy of the vast majority of the models created road safety performance indicators, researchers discussed the concept of 233
O
170 using this approach is that they depend upon total measures of move- benchmarking and have identified different domains and indicators of 234
171 ment speed (e.g., speed limit, travel speed) and volume (e.g., AADT or road safety. The concept of benchmarking was focused on a composite 235
172 hourly volumes) and, thus, are most certainly not adequate to distin- road safety index to identify the one value based road safety performance 236
C
173 guish the continuous blackspots (i.e., areas having a high likelihood of evaluation (Al Haji, 2005). For benchmarking and analysis, the following 237
174 accidents) made because of the cooperation of encompassing traffic five weighted method techniques have been applied with good 238
N
175 conditions with the geometric attributes of interstate portions (Abdel- results (Hermans, Brijs, Wets, & Vanhoof, 2009; Hermans, Van den 239
176 Aty & Pande, 2005). From previous studies it is evident that traffic engi- Bossche, & Wets, 2008) (factor analysis, analytic hierarchy process, 240
177 neers have always focused on studying the relationship of road crashes budget allocation, data envelopment analysis, and equal weighting; 241
U
178 (accidents and fatalities) (Dadashova et al., 2016; Golob et al., 2004; Hermans et al., 2008). Out of these five considered methods, data envel- 242
179 Golob et al., 2008; Imprialou et al., 2016; Lord, Manar, & Vizioli, 2005; opment analysis (DEA) is the most unique technique and provided a bet- 243
180 Ogwueleka et al., 2014; Pande & Abdel-Aty, 2006; Wang et al., 2009; ter understanding for the road safety performance index for ranking 244
181 Yasin Çodur & Tortum, 2015) in context of volume/capacity (Golob & purpose (Hermans et al., 2008; Hermans et al., 2009). 245
182 Recker, 2004; Lord, Manar, & Vizioli, 2005; Pande & Abdel-Aty, 2006;
183 Zhou & Sisiopiku, 1997), vehicles miles traveled (Abdel-Aty et al., 2.4. DEA for accident data analysis 246
184 2013; Dadashova et al., 2016; Frantzeskakis & Iordanis, 1987; Jovanis
185 & Chang, 1986; Wang & Kockelman, 2013), vehicles hours traveled As per the 2016 European Commission report, road safety perfor- 247
186 (Martin, 2002; Yeo et al., 2013; Zhou & Sisiopiku, 1997), speed (average mance analysis of motorways is necessary for the safety of travelers. 248
187 travel speed, posted speed, etc.) (Aljanahi et al., 1999; Elvik, 1997; Elvik In the last decade, studies have revolved around the application of 249
188 et al., 2004; Garber & Ehrhart, 2000; Golob & Recker, 2004; Imprialou DEA for benchmarking and safety analysis. Road safety performance 250
189 et al., 2016; Kononov et al., 2011; Malyshkina & Mannering, 2008; measurement is an important part of safety analysis and operations 251
190 Pande & Abdel-Aty, 2006), flow (vehicles/h) (Chimba & Kutela, n.d.; research. “DEA is a data-oriented method for measuring and 252
191 Aljanahi et al., 1999; Garber & Ehrhart, 2000; Golob et al., 2004; Golob benchmarking the relative performance of peer decision-making units 253
254 (DMUs) with multiple inputs and multiple outputs” (Alinezhad, 2016). outputs (i.e., risk evaluation for elaborating levels of severity and then 316
255 DEA was initiated in 1978 when Charnes, Cooper, and Rhodes (1978) risk prediction for factor impact analysis purpose; Shah et al., 2017). 317
256 established how to convert a fractional linear measure of efficiency (or Researchers use this combination of DEA and DT to evaluate risk and ef- 318
257 risk) into a linear programming format. This non-parametric approach ficiency in different fields like banking sectors (Alinezhad, 2016), tech- 319
258 solves a linear programming (LP) origination per DMU and the weights nology commercialization projects (Sohn & Moon, 2004), service units 320
259 allocated to each DMU are the result of the corresponding LP of firms (Seol, Choi, Park, & Park, 2007), Arabic banks (Samoilenko & 321
260 (Alinezhad, 2016; Amirteimoori & Kordrostami, 2012). DEA has been Osei-Bryson, 2013), hybrid supplier evaluation (Wu, 2009), telecom- 322
261 successfully applied in road safety research for the analysis of munication service (Samoilenko & Osei-Bryson, 2008), financial firms 323
262 European countries, and the concept of composite performance indica- (Lee, 2010), and insurance branches (Shafaghizadeh & Attari, 2014). 324
263 tor and DEA was used for weights determination and ranking. Road fa- This study will contribute with a successful applications of this 325
264 talities were tested on the basis of fatalities per million inhabitants DEA-DT joint approach, showing that it can be a useful technique for 326
265 (Hermans et al., 2008). To enhance the output of risk value, inputs like transportation engineers during road safety data analysis and decision 327
266 population, passenger kilometers, and number of registered cars were making process. 328
267 used against the output of a number of fatalities, but cluster analysis
F
268 technique was also added to analyze the performance of similar 3. Materials and methods 329
269 European countries (Shen, Hermans, Brijs, Wets, & Vanhoof, 2012). At
O
270 the municipality level in Israel, for the allocation of annual budget, 3.1. Study area and data description 330
271 DEA was used for number of drivers involved in crashes, number of
272 seatbelt tickets issued per year, the percentage of cars, and the average Belgium is located in the heart of the European Schengen area. From 331
O
273 age of private cars were (Alper, Sinuany-Stern, & Shinar, 2015). Simi- European Commission reports (CARE, 2016; EU, 2016), the problematic 332
274 larly, for road safety performance analysis, member states of Brazil sector identified for analysis is based on these reports as motorways are 333
275 334
R
used DEA, mortality rate, and fatality rate (fatality per vehicle & fatality most problematic areas in Belgium with respect to safety issues. The
276 per vehicle kilometer traveled) as output variables (Bastos et al., 2015). study area selected for this study is two Motorways in Belgium named 335
277 For road safety analysis of Serbia, DEA method was applied to data for 27 E-313(57 Km) and E-314(50 Km) (Limburg Province Sections). For 336
P
278 police departments for traffic and public risk evaluation (Rosić, Pešić, the selected Motorways (E-313 & 314), 67 segments were having at 337
279 Kukić, Antić, & Božović, 2017) and road safety risk for motorways in least one crash in 3 years. So data have been derived from the Flemish 338
280 Belgium (Shah et al., 2017). So from the discussion about DEA applica- Ministry of Mobility and Public Works and FEATHERS model
D 339
281 tion in the field of transportation engineering, it confirms that DEA is (Janssens, Wets, Timmermans, & Arentze, 2007) for these motorways 340
282 one of the established techniques to evaluate the severity level of road as shown in Fig. 3. 341
283 safety (Shah et al., 2017).
E
3.2. Framework for decision-making process 342
284 2.5. DT for accident data analysis
T
To evaluate the road safety condition of roads, a two-stage perfor- 343

285 Decision trees as a data mining technique allows DMUs to discover mance analysis concept was introduced, as shown in Fig. 4. This method 344
C
286 significant information that had previously been hidden in large data- consists of two sections; first DEA model portion to evaluate the severity 345
287 bases. It was developed in 1984 by Breiman (Breiman, Friedman, level of accident locations, and second the decision tree mechanism to 346
288 Stone, & Olshen, 1984) and improved in 1996 by Ripely (Miller & evaluate the impact of the factors on the severity. It also helps decision 347
E
289 Ripley, 1996). The problem is illustrated by a decision making tree so makers (engineers) to locate the location and identify attributes that 348
290 that each non-leaf node is associated with one of the decisions making need to be improved during decision making and evaluation. Process 349
R
291 variables, each branch of a non-leaf node is associated with a subset of of application of DEA in collaboration with DT model to calculate the 350
292 the decision making variables values, and each leaf node is linked to a Risk level of road segments can be explained as. 351
293 target variable (the dependent variable) value (Rahimi, Behmanesh, &
R
294 Yusuff, 2013). Each leaf is associated with a target variable's mean Step 1: Selection of the problematic motorways or highways (Limburg- 352
295 value; therefore, this tree can be an alternative to continuous linear Motorways E-313 & E314). 353
O
296 models for solving the problems of regression analyses of classified Step 2: Segmentation of the selected motorways according to the data 354
297 data (Clark, 1992). Researchers have vastly used this reliable technique network. 355
298 in road safety field and found relationship significance of crashes and Step 3: Selection of the problematic section having at least one Accident 356
C
299 factors like geometric pattern (Karlaftis & Golias, 2002), collision order in three years. 357
300 in Korea (Sohn & Lee, 2003), geometric factors (Chang & Chen, 2005), Step 4: Selection of parameters from accident data (no. of accidents & 358
N
301 causality class (Kashani & Mohaymany, 2011), crash seriousness no. of persons injured or killed), traffic features data (VMT, 359
302 (Abellán, López, & De OñA, 2013), geometric factors along with number VHT,V/C, speed and flow), and road feature data (horizontal 360
303 of vehicles (Chang & Chien, 2013), and over speeding (Olutayo & curve, vertical curve) with reference to domain of transportation 361
U
304 Eludire, 2014). From the analysis, we can summarize that DT was previ- engineers. 362
305 ously used as an road safety analysis tool, which is a useful technique to Step 5: Data were distributed and applied into two section: 363
306 study the impact of road-related features, geometry, and other contrib- Section I: DEA Models to identify risk level and rank priority wise 364
307 uting factors on road safety. most risky road segments with reference to road safety. Following the 365
basic concept of risk 366
308 2.6. DEA-DT approach for accident data analysis
Road Safety Outcome
Risk ¼ ð1Þ
309 A combination of DEA and DT has not been used in the road safety Exposure
310 field; however, it is a popular method in other fields, like banking and
311 corporate sectors. Previous studies concluded that DEA is a suitable 368
Q11
312 tool for risk calculation, but for risk prediction and decision-making pro- While calculating road safety risk, road safety outcome is the num-
313 cess with reference to contributing factors, decision tree does not fit, so ber of accidents and number of persons injured or killed (NoA and 369
314 a discussion started after (Athanassopoulos & Curram, 1996) on com- NoAP), while exposure variables are related to traffic (i.e., VMT, V/C 370
315 bining DEA with a data mining technique to get best the possible and VHT). 371
F
O
O
R
P
Fig. 3. Study area of Limburg.
D
372 Step 6: After calculating risk value using linear program with the 2012). Researchers have explained DEA as “Consider an n-DMUs set, 407
373 help of Lingo software, some values were the same (i.e., 1) and it was each consuming m different inputs to produce s different outputs. The 408
E
374 difficult to rank them. Therefore, a cross-efficient approach was applied relative efficiency of a DMU is defined as the ratio of its total weighted 409
375 that follows the concept of using the weights of each DMU by multiply- output to its total weighted input, subjected to lie between zero and 410
T
376 ing it with the values variables of each DMU to calculate a unique value the unity. Mathematically, the efficiency score of a particular DMU0, 411
377 called a cross-efficiency method and the value obtained is named as Risk i.e., E0, is obtained by solving the following constrained optimization 412
C
378 CE. Ranking helps in prioritizing the risky segments to set the priority of problem (Charnes et al., 1978; Shen et al., 2012): 413
379 risky segments.
Ps
380 Step 7: Development of GIS maps to identify and facilitate the trans- ur yr0
E
E0 ¼ max Pr¼1
m
381 portation engineers for decision-making procedure. i¼1 vi xi0 415
382 Step 8: DT models to identify the factors influencing Risk level (iden-
416
R
383 tified by DEA model) of each segments to mitigate and control contrib- Ps
ur yrj
384 uting factors to control and improve the road safety situation. subject to Pr¼1
m ≤1; j ¼ 1; …:; n
i¼1 vi xij 418
385 Step 9: Rearrangement of data for application of DT model (i.e., To
R
386 normalize the data of risk value nature log has been applied so named 419
387 as NLog Risk and it was considered as dependent variables while ur ; vi ≥0; r ¼ 1; …::; s; i ¼ 1; …; m ð2Þ
O
388 speed, flow, horizontal curve and vertical curve were considered as in-
389 dependent variables). where yrj and xij are the r-th output and i-th input, respectively of the 421
Q7
390 Step 10: A comparative performance of DT in comparison with MLR j-th country, ur is the weight given to output r, and vi is the weight
C
391 (Multiple Linear Regression) was analyzed by comparing R2 (Coefficient given to input i.” 422
392 of Determination) and RMSE (Root mean squared error) value. The concept of DEA model has been shown in Fig. 5. “It is a linear 423
programming technique for measuring the relative performance of en- 424
N
393 Step 9: Analysis of factors influencing the risk value of road segments
394 as a single unit and in combination of other factors. tities or units of a similar pattern. Highway sections are considered to 425
395 Step 10: Discussion on obtained results as policymakers and trans- be Decision-Making Units (DMUs) for the application of the DEA 426
U
396 portation engineers. model, and the risk level is calculated by applying the road safety out- 427
come and exposure variables” (Shah et al., 2018). 428
397 4. Analysis and discussion During the severity level risk analysis it can be assumed that “the 429
lowest level has been considered as the frontier of safety. In the case 430
398 4.1. Section-I: decision making regarding road sections of efficiency evaluation, Efficiency is calculated by maximizing output 431
and minimizing input, while, for calculating risk, we minimize the 432
399 During the analysis phase-I, four major steps are conducted to eval- output and maximize the input (Shah et al., 2018). The simplest form 433
400 uate road safety risk value: (i) Selection of input and output variables; of calculating Efficiency by DEA is as follows” (Shah et al., 2018). 434
401 (ii) application of DEA model for calculation of calculation of risk; (iii) Efficiency: The basic concept of the DEA efficiency calculation is as in 435
402 calculation of risk using cross efficient matrix (called Risk CE) for rank- Eq. (3) (Shah et al., 2018). 436
403 ing of risk value for prioritization of risky segments; and (iv) normaliza-
404 tion of risk value by taking Natural Log (called NLog Risk) to be used in Weighted Sum of Output Maximize Output
Efficiency ¼ ¼ ð3Þ
405 nest phase. So, road traffic crashes and fatalities have been taken as out- Weighted Sum of Input Minimize Input
406 put while exposure variables have been taken as input (Shen et al., 438
Following the above-explained concept, risk is calculated as in DMUs (Dyson & Thanassoulis, 1988; Wong & Beasley, 1990). Sometime 450
439 Eq. (4): due to similar efficiency or risk value, ranking procedure is difficult 451
to set priorities of DMUs. To overcome these difficulties, a cross- 452
Weighted Sum of Output Minimize Output efficiency method (Sexton, Silkman, & Hogan, 1986) was developed as 453
Risk ¼ ¼
Weighted Sum of Input Maximize Input a DEA extension tool that can be used to identify the best overall per- 454
ð4Þ
Road Safety Outcome formers and to effectively rank all DMUs. The main theme of this con- 455
¼
Exposure cept is use weights of each DMU to evaluate efficiency and a sum is 456
441 used to get the best possible value. The conceptual demonstration can 457
“DEA model is applied through Lingo software (programming- be seen in cross-efficiency matrix (CEM) as shown in Table 2. In the de- 458
442 based), which provides the severity values for each segment of road veloped matrix, the values in the i-th row and j-th column interpret the 459
443 (i.e., Decision-Making Unit-DMU). Researchers in the field of road safety worst or best scores of DMU j using the optimal weights of DMU i. The 460
444 introduced DEA by assigning weights for the construction of composite basic DEA efficient or risky values are thus kept in the major diagonal. 461
445 performance indicators; then, for the evaluation of road safety rankings, Each column of the matrix is then averaged to obtain a mean cross- 462
446 a risk value was calculated” (Shah et al., 2018). efficiency or risky score for each DMU. This process can help in selecting 463
F
447 DEA holds an smart feature that each DMU is permitted to select highest to lowest efficiency for ranking purpose (Boussofiane, Dyson, & 464
448 its own most promising input and output weights, or multipliers for Thanassoulis, 1991). Therefore, this process can be considered as a type 465
O
449 evaluating its best efficiency, rather than the same weights for all the of sensitivity analysis since specific sets of weights are applied to each 466
O
R
P
D
E
T
C
E
R
R
O
C
N
U
Fig. 4. DEA-DT framework for accident risk evaluation.
Table 2 t2:1
A generalized cross-efficiency matrix (CEM). Q1
t2:2
Rating DMU Rated DMU t2:3
1 2 3 … n t2:4
1 E11 E12 E13 … E1n t2:5

2 E21 E22 E23 … E2n t2:6
3 E31 E32 E33 … E3n t2:7
. . . . . . t2:8
. . . . . . t2:9
. . . . . . t2:10
n En1 En2 En3 … Enn t2:11
Mean E1 E2 E3 … En t2:12
and a chain of steps can be decided to evaluate the severity and risk 490
F
level for certain section of highways. 491
O
4.2. Section-2: decision making regarding factors 492
Although the identification of the worst section on the basis of safety 493
O
Fig. 5. DEA conceptual model.
issues is one of the key tasks during safety audit, safety can be ensured 494
by improving the factors contributing to a safety problem. Decision 495
R
467 unit or segment, and they are all internally calculated rather than exter- Makers/Transportation Engineers need to know the basic contributing 496
468 nally provided. This concept has been explained as it is by (Shen et al., factors. Decision tree technique has been applied to analyze the impact 497
469 2012). of contributing factors (Description of factors can be observed in 498
P
470 In this study, the concept of identification of risk was elaborated as Table 3.). 499
471 road safety outcome, which was basically associated with road acci- Traffic flow characteristics and geometric design features are major 500
472 dents and causalities. However, the exposure was defined as Volume/ factors to be dealt with by engineers in order to improve the safety con-
D 501
473 Capacity, Vehicle Miles Traveled, and Vehicle Hourly Traveled (Shah dition of a highways or road. Therefore, these four factors shouldbe con- 502
474 et al., 2017) as shown in Table 1. It is one of the familiar concepts by sidered during the decision making process. These factors can be 503
475 the researchers based on Model (1), a segment with a value of up to 1 improved and a budget can be allocated for the improvement of these 504
E
476 is considered as the safest one while the Highway segment with the factors during the budget distribution from the highway agency and 505
477 maximum value is considered as the most dangerous one (Shah et al., federal government. In choice tree displaying, an exact tree speaks to 506
T
478 2017). Moreover, by using the weights calculated for each factor are a division of the information that is made by applying a progression of 507
479 used to calculate risk by the cross risk method (Shen et al., 2012) so straightforward guidelines. In the creation of a decision tree, a data set 508
C
480
Q8 that all the DMUs can be made comparable. (See Table 6.) is usually divided into two parts: the training data set and the test 509
481 GIS mapping is one of the advanced tools to use to visualize the con- dataset (Hand, Mannila, & Smyth, 2001; Jiawei & Kamber, 2001). 510
482 cept of the analysis and it provides an opportunity to evaluate and ana- They then undergo two main processing phases of growth and prun- 511
E
483 lyze the big data in layers. The joining of risk value and severity ing. In the development stage, a decision tree is constructed from a set of 512
484 calculation by DEA model can be visualized by GIS (Shah et al., 2017). training data. In this phase, each leaf node is associated with a class. A 513
R
485 Previously for the same section analysis has been done to visualize the noteworthy favorable position of the choice tree over other displaying 514
486 risk as shown in Fig. 6. Risk value evaluated through DEA has been visu- systems is that it delivers a model that may speak to interpretable prin- 515
487 alized in combination with GIS in previous researches (Shah et al., 2017) ciples or rational explanations. After calculation of risk evaluation, DT 516
R
488 a step further, the decision tree mechanism has been employed to get a technique was applied to check which factors influence the risk factor 517
489 view that these results can be focused on the basis of tree mechanism of DMUs, as shown in Fig. 7. The performance of a model is predicted 518
O
t1:1 Table 1
C
t1:2 Data description for motorways of Belgium Limburg section.
t1:3 Road Variable Units N Mean SD Min. Med. Max.

N
t1:4 E-314 Traffic Flow Veh/h 36 1116.4 414.2 44.8 1272.7 1483.4
t1:5 Travel Speed Mile/h 36 107.82 7.34 99.51 104.86 120
t1:6 Horz_Curve 1-Yes, 0-No 36 0.7778 0.4216 0 1 1
U
t1:7 Vert_Curve 1-Up, 2-Down, 3-No 36 1.917 0.732 1 2 3

t1:8 V/C Volume/capacity 36 0.5131 0.1442 0.139 0.581 0.62
t1:9 VMT Veh. mile travel 36 1718 1154 77 1567 4366
t1:10 VHT Veh. hr travel 36 1061 737 38 1014 2986
t1:11 Segment Length Km 36 1.068 0.736 0.09 0.955 3.35
t1:12 NoA Numb. of accidents 36 10.92 11.71 1 5 49
t1:13 NoAP Numb. of affected persons (inj/killed) 36 16.56 18.09 1 8 76
t1:14 E-313 Traffic flow Veh/h 31 795.9 433.1 31.4 794.4 1330
t1:15 Travel speed Mile/h 31 114.66 7.76 96.89 119.44 120
t1:16 Horz_Curve 1-Yes, 0-No 31 0.8387 0.3739 0 1 1
t1:17 Vert_Curve 1-Up, 2-Down, 3-No 31 2.032 0.706 1 2 3
t1:18 V/C Vol/capacity 31 0.3562 0.1842 0.08 0.327 0.643
t1:19 VMT Veh. mile travel 31 1955 1630 83 1668 5186
t1:20 VHT Veh. hr travel 31 1129 1031 41 954 3616
t1:21 Segment length Km 31 1.518 0.965 0.22 1.63 3.77
t1:22 NoA Numb. of accidents 31 8.03 14.64 1 3 74
t1:23 NoAP Numb. of affected persons (inj/killed) 31 11.81 21.13 1 6 105
t3:1 Table 3
t3:2 Application of DEA model for accident analysis of motorway segments.
t3:3 DMU Input 1 Input 2 Input 3 Output 1 Output 2 DEA Risk DEA CE Risk NLog_Risk RANK
t3:4 Seg_ID V/C VMT VHT RA NoAP
t3:5 1 0.369 3039.22 1541.61 74 105 60.79 91.07 4.51 1

t3:6 29 0.139 109.17 54.58 6 8 25.72 71.73 4.27 2
t3:7 19 0.603 183.73 118.16 12 20 11.95 69.92 4.25 3
t3:8 2 0.384 2494.33 1268.38 49 76 51.90 65.10 4.18 4
t3:9 34 0.080 82.51 41.26 3 6 22.62 62.90 4.14 5
t3:10 5 0.278 2190.90 1096.68 38 50 41.09 62.28 4.13 6
t3:11 25 0.139 76.74 38.37 3 6 13.02 58.11 4.06 7
t3:12 26 0.237 202.31 101.24 9 11 22.58 57.15 4.05 8
t3:13 3 0.361 2683.94 1361.27 40 61 39.16 53.26 3.98 9
t3:14 21 0.534 594.78 336.96 13 24 14.65 35.40 3.57 10
t3:15 – – – – – – – – – –
t3:16 – – – – – – – – – –
F
t3:17 27 0.499 4147.38 2297.64 4 6 2.50 3.61 1.28 51
t3:18 33 0.486 3158.98 1742.42 3 7 3.05 3.50 1.25 52
O
t3:19 36 0.585 5013.80 3103.21 4 8 2.30 3.19 1.16 53
t3:20 59 0.337 1882.65 954.05 2 2 2.02 2.92 1.07 54
t3:21 65 0.439 441.58 226.31 1 1 1.35 2.79 1.03 55
t3:22 55 0.602 1711.50 1152.94 2 3 1.91 2.34 0.85 56
O
t3:23 52 0.595 3400.89 2256.01 3 3 1.46 2.33 0.85 57
t3:24 47 0.603 1435.26 944.89 2 2 1.94 2.26 0.82 58
t3:25 51 0.465 618.77 328.33 1 1 1.27 2.17 0.77 59
R
t3:26 61 0.575 808.56 517.91 1 2 1.05 2.16 0.77 60
t3:27 60 0.555 2423.61 1428.66 2 2 1.52 1.90 0.64 61
t3:28 48 0.486 3956.90 2182.53 2 2 1.00 1.73 0.55 62
P
t3:29 46 0.603 1159.82 763.55 1 2 1.00 1.66 0.51 63
t3:30 53 0.631 4275.09 3046.69 2 3 1.00 1.47 0.39 64
t3:31 67 0.592 1093.97 734.83 1 1 1.00 1.31 0.27 65
t3:32 49 0.499 3214.27 1780.70 1 2 D 1.00 1.08 0.08 66
t3:33 66 0.575 1714.22 1098.00 1 1 1.00 1.07 0.07 67
t3:34 Note: DEA Risk: Risk value calculated by Data Envelopment Analysis(DEA) Method.
t3:35 DEA Risk CE: Cross Efficient Risk calculated after application of DEA to get the unique Risk value for ranking.
E
t3:36 Ln(Risk)-Natural Log of Risk value-To Normalize the Data.
T
519 by using the basic parameters of statistics. These parmeters are root The proposed decision tree in this study includes four main compo- 533
C
520 mean squared error (RMSE), which is the difference between the actual nents: The first component is the output (dependent) variable. Based on 534
521 and the predicted values and coefficient of determination (R2) the independent (predictive) variables, this variable is used to predict. 535
522 (Siddique, Aggarwal, & Aggarwal, 2011). This tree diagram shows that speed is a major factor affecting risk eval- 536
E
uation of each DMU, as shown in Table 3. Thus, as a traffic engineer, one 537
rffiffiffiffiffiffi N can analyze which factor is most important and what can be controlled 538
1X
R
2
RMSE ¼ ð ðactual−predictedÞ Þ ð5Þ by changing the speed limit of those segments, creating a safety zone 539
N n¼1 with this low-cost solution. 540
524
A leaf table report as shown in Table 4 has been developed to show 541
R
525
SSE the possible combinations (leaf 5,6,7,8,9,10 and 11 are associated with 542
R2 ¼ 1− ð6Þ high-risk levels). With the help of this technique, a targeted treatment 543
SSy
O
can opt for all road segments. 544

527
SSE is the sum of squared errors of prediction and SSy is total varia-
C
528 tion. Mean absolute error is similar to root mean square, except using Table 5 t5:1
529 absolute difference instead of squared difference. The model after devel- Parametric estimates of decision tree model. t5:2
opment are compared by coefficient of determination (R2). A classic fit
N
530
Impact of factors (independent variables) t5:3
531 would bring about a R2 of 1, and poor fit almost 0 while RMSE should
532 be as minimum as possible (Siddique et al., 2011). Term No. of SS Portion t5:4
U
splits
SPEED 1 25.04 0.4844 t5:5

t4:1 Table 4 FLOW 6 16.33 0.3158 t5:6
t4:2 Descriptive statistics for phase-II analysis. Vert_Curve 1 7.80 0.1508 t5:7
Horz_Curve 2 2.53 0.0489 t5:8
t4:3 Variables Description Mean S.D Min. Max.
Cross-validation Value t5:9
t4:4 Ln(Risk) Risk-severity value of the section 2.212 1.220 0.068 4.512 R2 Actual 0.52 t5:10
(dea-based) K-Folded 0.32 t5:11
t4:5 Flow Average annual daily traffic on each 968.1 449.6 31.5 1483.4 RMSE 0.83 t5:12
segment – vph t5:13
t4:6 Speed Average travel speed for each 110.99 8.23 96.89 120 Model performance: comparative analysis of DT Vs MLR t5:14
segment – kph Model R2 R2 RMSE Remarks t5:15
t4:7 Horz_Curve 1-Presence of curve, 0-Tangent 0 1 (Main) (K-Fold)
t4:8 Vert_Curve 1-Curve upward, 2-Curve 1 3 Prd. Val.
downward, 3-Flat DT 0.52 0.31 0.83 Better t5:16
MLR 0.27 0.14 1.07 Model t5:17
t4:9 Note: Ln(Risk) – Natural Log of Risk value – To Normalize the Data.
t6:1 Table 6
t6:2 Leaf table showing the relationship between risk level and factors.
t6:3 Leaf label Leaf Mean Count
t6:4 SPEED b118.86&Vert_Curve b3&FLOW≥ 1197.07&FLOWb1276.10 1 0.66304286 7

t6:5 SPEED b118.86&Vert_Curve b3&FLOW≥ 1197.07&FLOW≥ 1276.10&Horz_Curveb1 2 1.2646 5
t6:6 SPEED b118.86&Vert_Curve b3&FLOW≥ 1197.07&FLOW≥ 1276.10&Horz_Curve≥ 1&FLOW≥ 1436.09 3 1.30098333 6
t6:7 SPEED b118.86&Vert_Curve b3&FLOW≥ 1197.07&FLOW≥ 1276.10&Horz_Curve≥ 1&FLOWb1436.09 4 1.86458 5
t6:8 SPEED b118.86&Vert_Curve b3&FLOWb1197.07 5 2.24734286 7
t6:9 SPEED b118.86&Vert_Curve ≥ 3&Horz_Curve≥ 1 6 2.03361667 6
t6:10 SPEED b118.86&Vert_Curve ≥ 3&Horz_Curveb1 7 2.93926 5
t6:11 SPEED ≥ 118.86&FLOW≥ 171.83&FLOWb794.44&FLOW≥ 502.50 8 2.33522222 9
t6:12 SPEED ≥ 118.86&FLOW≥ 171.83&FLOWb794.44&FLOWb502.50 9 2.9799 5
t6:13 SPEED ≥ 118.86&FLOW≥ 171.83&FLOW≥ 794.44 10 3.35491667 6
t6:14 SPEED ≥ 118.86&FLOWb171.83 11 3.57085 6
F
545 From Tables 4 and 5 and Fig. 7 we can conclude that the Decision were of major concern. After effectively applying the DEA-DT approach 557
O
546 Tree model is presenting results showing that speed and flow are for motorways safety assessment, we can likewise concentrate on the 558
547 influencing factors. Thus in the decision-making process for an en- contributing variables utilized as a part of the hazard expectation. Traffic 559
548 gineer, it is a useful method to identify the major effecting factor engineers first like to concentrate on contributing factors, which are of 560
O
549 (i.e., speed factor). The model performance and validation low cost. In this manner from a graphical examination of the contribut- 561
550 (K-folded method) has been tested through the basic tests like R2 ing variables, we can see that the dominant part of the crashes is on the 562
R
551 (0.5) and RMSE(0.84), which show the predictive strength of risky segments of the motorways. Safety engineers, as a rule, abstain 563
552 applied DT model in comparison to MLR as shown in Table 4 and from going for an infrastructural change in light of the fact that updating 564
553 Fig. 8. and recreation is an expensive strategy, so in the event that they con- 565
P
centrate on the ease treatment, can go for speed and flow control. 566
554 4.3. Analysis of factors Fig. 9 shows the connection between the hazard and the diverse con- 567
tributing components.
D 568
555 GIS-based mapping procedure guides decision makers up to the There are four major independent variables that influence risk value 569
556 identification of the problematic areas but still, the improvement factors of motorway segments: 570
E
T
C
E
R
R
O
C
N
U
Fig. 6. Accident severity risk map for motorway segments.
F
O
O
R
P
Fig. 7. Road safety risk prediction through decision tree model.

D
E
T
C
E
R
R
O
C
N
U
F
O
O
Fig. 8. Performance analysis of risk prediction through DT and MLR model.
R
571 4.3.1. Speed horizontal curve design is a key prospect to watch during the safe 598
572 Speed is considered one of the most important factors influencing road design. Researchers explained that accidents on curves are three 599
573 road safety. One can easily understand, according to the principles of times higher than that of straight road sections (Torbic et al., 2004). 600
P
574 physics, that if an object hits another object with higher speed than Moreover, there is a serious concern by decision makers for accidents 601
575 the impact of that crash/collision will be severe. In the field of highway on horizontal curves (Bonneson, Pratt, Miles, & Carlson, 2007). Motor- 602
576 engineering, one major necessity is to enhance a smooth maneuvering vehicles crashes occur frequently at horizontal curves (Persaud,
D 603
577 environment for road users. Particularly in the case of motorways, de- Retting, & Lyon, 2000). Horizontal curves contribute frequently in 604
578 signers prefer to maintain a higher speed limit to provide road users a accidents because it causes safety hazard to road users with reference 605
579 speedy and comfortable environment for time saving and smooth trav- to drivers prospective (Schneider IV, Zimmerman, Van Boxel, & 606
E
580 eling. Sometimes this compromises safety, as with high speed there are Vavilikolanu, 2009). Change in design of horizontal curve reduces the 607
581 other factors that combine, causing high number of accidents. In the sight distance prospective and effects vehicular handling capabilities 608
T
582 case of motorways (E-313 & E-314), high speed is a contributing factor: (Schneider IV, Savolainen, & Moore, 2010). During the analysis of Mo- 609
583 above 115 km/h shows a higher risk factor, so reduction in the speed torways of Limburg, there was no problem found due to horizontal 610
C
584 limit from 120 to 110 km/h can improve the safety of these motorways. curve design. 611
4.3.4. Vertical curve 612

E
585 4.3.2. Flow

The second factor of geometric design is vertical curve. Sometimes 613
586 It is a common understanding that increasing the number of users on
vertical curve also contributes to accidents. However, it usually occurs 614
587 the road will contribute to the increasing number of accidents. Flow is
R
in hilly areas and in this section of Limburg motorways there was no 615
588 considered as the number of vehicles/h, which is a major factor because
issue relevant to this type of design features, so motorways are safe re- 616
589 increased population also contributes in occupying road space. The be-
garding vertical curve perspective. 617
R
590 havior variation of vehicles also increases with the increase in vehicles,
591 which contribute in risk increment. Higher traffic flow creates an envi-
4.3.5. Combined effect of variables 618
592 ronment of congestion, which also contributes to increased road acci-
O
Sometimes a single factor does not affect safety, but a combination of 619
593 dents. This is not a singular influence on road safety, but it works in
factors affects the safety performance of roads. Therefore, to analyze the 620
594 combination with other factors like speed and geometric design.
risk level, a combination effect of different variables has been studied. 621
C
NLog Risk value is natural log value of risk, which is a composite factor 622
595 4.3.3. Horizontal curve indicating the safety level of a road. Risk level has been used to analyze 623
N
596 Geometric design is one of the major components of road design and highly risky situations under all four contributing factors. During 624
597 a true contributor in road safety. As a transportation engineer, the analysis of flow and speed combination impact on road safety 625
U
Fig. 9. Impact analysis of factors on risk-severity.
F
O
O
R
P
D
Fig. 10. Tree graph distribution for contributing factor based risk analysis.
E
626 performance, we observe that increases in speed and higher flow con- safety hazard with the assistance of the DT method, the following com- 659
627 tribute to an increased risk value. Flow in combination with horizontal ponents were considered: speed, flow, even, and vertical curve. They are 660
T
628 curve shows no impact or increase in risk value. Flow in combination the most imperative variables affected by the safety engineer. Generally, 661
629 with vertical curve shows no trend of increasing risk level. So, there is an infrastructural change like reworking the horizontal and vertical 662
C
630 no problem of geometric design. The tree graph shows the actual loca- curve can be costly. Yet, this framework can help to pinpoint problem 663
631 tion of risk under all contributing factors. We can see from Fig. 10 that areas on the interstate (i.e., especially in case of 100 km long roadway), 664
E
632 the hazard level can be decreased by controlling two elements and tackling the safety issue of just those portions. By combining DEA 665
633 (i.e., speed and flow). with DT, GIS in a road safety investigation framework can also incorpo- 666
634 We can see from the information that, for example, 35 of 67 sections rate the usefulness of the DEA and, in the meantime, enhance the ar- 667
R
635 with 52% are over the mean speed constrain 110 km/h in parallel to traf- rangement of potential utilization of GIS. The framework is adaptable 668
636 fic flow above mean level of movement stream. By controlling these two and self-versatile, and equipped for consolidating any change in new in- 669
R
637 variables we can improve safety on the motorways with a low-cost formational collection. Therefore, a joint approach of DEA-DT can give a 670
638 treatment. We can also observe that if we target a higher level of safety, simple and proficient yield for data analysts for road safety accident 671
639 the majority of accidents are on flat vertical curves, so geometric point analysis and decision making for allocating road infrastructure improve- 672
O
640 of view is clear. ment budgets. 673

C
641 5. Conclusions Acknowledgments 674
This research is jointly supported by TITE and IMOB for publication. 675
N
642 This study focuses on accident analysis of road infrastructure geom-

643 etry and traffic stream characteristics. To improve the estimation preci- Authors would like to thank HE-Boong Kwon (USA), one of the pioneers 676
644 sion, a two-phase framework was proposed to accomplish the hazard of a similar method for his valuable guidance. 677
Q9
U
645 assessment (i.e., a benchmarking instrument of DEA combined with a

646 forecast model of DT was introduced to road safety research). A crash References 678
647 dataset from the Flemish Road safety was stratified in two elements: Abdel-Aty, M., Lee, J., Siddiqui, C., & Choi, K. (2013). Geographical unit based analysis in 679
648 number of accidents and number of influenced people (killed or in- the context of transportation safety planning. Transportation research part A: Policy 680
649 jured) (Shah et al., 2017), and was used to display the proposed devel- and practice. Vol. 49. (pp. 62–75). 681
Abdel-Aty, M., & Pande, A. (2005). Identifying crash propensity using specific traffic speed 682
650 opment of DEA and DT regression. Decision trees as a technique of conditions. Journal of Safety Research, 36(1), 97–108. 683
651 controlled data mining are a classification method that is convenient Abdel-Aty, M., & Pande, A. (2007). Crash data analysis: Collective vs. individual crash level 684
652 to interpret and understand when compared to other performance approach. Journal of Safety Research, 38(5), 581–587. 685
653 models. In the first stage, DEA helps to identify the risky segments of Abellán, J., López, G., & De OñA, J. (2013). Analysis of traffic accident severity using deci- 686
sion rules via decision trees. Expert Systems with Applications, 40(15), 6047–6054. 687
654 the motorways and in the second stage contributing factors were iden- Aguero-Valverde, J. (2013). Full Bayes Poisson gamma, Poisson lognormal, and zero in- 688
655 tified by DT. According to the analysis, targeting speed limit is one of the flated random effects models: Comparing the precision of crash frequency estimates. 689
656 major low-cost treatments and provides that, by reducing speed limit, Accident Analysis & Prevention, 50, 289–297. 690
Al Haji, G. (2005). Towards a road safety development index (RSDI): Development of an 691
657 we can reduce major risk; controlling flow level by controlling access international index to measure road safety performance. Linköping University Elec- 692
658 to the motorways can also reduce risk. To check the consistency of a tronic Press. 693
694 Al-Ghamdi, A. S. (2002). Using logistic regression to estimate the influence of accident Geurts, K., Wets, G., Brijs, T., & Vanhoof, K. (2004). Identification and ranking of 777
695 factors on accident severity. Accident Analysis & Prevention, 34(6), 729–741. black spots: Sensitivity analysis. Transportation Research Record: Journal of the 778
696 Alinezhad, A. (2016). An integrated DEA and data mining approach for performance as- Transportation Research Board(1897), 34–42. 779
697 sessment. Iranian Journal of Optimization, 8(2), 968–987. Givehchi, S., Hemmativaghef, E., & Hoveidi, H. (2017). Association between safety leading 780
698 Aljanahi, A., Rhodes, A., & Metcalfe, A. V. (1999). Speed, speed limits and road traffic acci- indicators and safety climate levels. Journal of Safety Research, 62, 23–32. 781
699 dents under free flow conditions. Accident Analysis & Prevention, 31(1), 161–168. Golob, T. F., Recker, W., & Pavlis, Y. (2008). Probabilistic models of freeway safety perfor- 782
700 Alper, D., Sinuany-Stern, Z., & Shinar, D. (2015). Evaluating the efficiency of local munic- mance using traffic flow data as predictors. Safety Science, 46(9), 1306–1333. 783
701 ipalities in providing traffic safety using the Data Envelopment Analysis. Accident Golob, T. F., & Recker, W. W. (2003). Relationships among urban freeway accidents, traffic 784
702 Analysis & Prevention, 78, 39–50. flow, weather, and lighting conditions. Journal of Transportation Engineering, 129(4), 785
703 Amirteimoori, A., & Kordrostami, S. (2012). A distance-based measure of super efficiency 342–353. 786
704 in data envelopment analysis: An application to gas companies. Journal of Global Golob, T. F., & Recker, W. W. (2004). A method for relating type of crash to traffic flow 787
705 Optimization, 54(1), 117–128. characteristics on urban freeways. Transportation Research Part A: Policy and 788
706 Anastasopoulos, P. C. (2016). Random parameters multivariate tobit and zero-inflated Practice, 38(1), 53–80. 789
707 count data models: Addressing unobserved and zero-state heterogeneity in accident Golob, T. F., Recker, W. W., & Alvarez, V. M. (2004). Tool to evaluate safety effects of 790
708 injury-severity rate and frequency analysis. Analytic Methods in Accident Research, 11, changes in freeway traffic flow. Journal of Transportation Engineering, 130(2), 791
709 17–32. 222–230. 792
710 Anastasopoulos, P. C., Mannering, F. L., Shankar, V. N., & Haddock, J. E. (2012). A study of Haddon Jr, W. (1980). Advances in the epidemiology of injuries as a basis for public pol- 793
711 factors affecting highway accident rates using the random-parameters tobit model. icy. Public Health Reports, 95(5), 411. 794
F
712 Accident Analysis & Prevention, 45, 628–633. Hakkert, A., Gitelman, V., & Vis, M. (2007). Road safety performance indicators: Theory. 795
713 Anastasopoulos, P. C., Shankar, V. N., Haddock, J. E., & Mannering, F. L. (2012). A multivar- Deliverable D3, 6. 796
714 iate tobit analysis of highway accident-injury-severity rates. Accident Analysis & Hamzeie, R., Savolainen, P. T., & Gates, T. J. (2017). Driver speed selection and crash risk: 797
O
715 Prevention, 45, 110–119. Insights from the naturalistic driving study. Journal of Safety Research, 63, 187–194. 798
716 Anastasopoulos, P. C., Tarko, A. P., & Mannering, F. L. (2008). Tobit analysis of Hand, D. J., Mannila, H., & Smyth, P. (2001). Principles of data mining (adaptive computation 799
717 vehicle accident rates on interstate highways. Accident Analysis & Prevention, 40(2), and machine learning). Cambridge, MA: MIT Press. 800
718 768–775. Hermans, E., Brijs, T., Wets, G., & Vanhoof, K. (2009). Benchmarking road safety: Lessons 801
O
719 Athanassopoulos, A. D., & Curram, S. P. (1996). A comparison of data envelopment anal- to learn from a data envelopment analysis. Accident Analysis & Prevention, 41(1), 802
720 ysis and artificial neural networks as tools for assessing the efficiency of decision 174–182. 803
721 making units. Journal of the Operational Research Society, 1000–1016. Hermans, E., Van den Bossche, F., & Wets, G. (2008). Combining road safety information in 804
R
722 Bastos, J. T., Shen, Y., Hermans, E., Brijs, T., Wets, G., & Ferraz, A. C. P. (2015). Traffic fatality a performance index. Accident Analysis & Prevention, 40(4), 1337–1344. 805
723 indicators in Brazil: State diagnosis based on data envelopment analysis research. Imprialou, M. -I. M., Quddus, M., Pitfield, D. E., & Lord, D. (2016). Re-visiting crash–speed 806
724 Accident Analysis & Prevention, 81, 61–73. relationships: A new perspective in crash modelling. Accident Analysis & Prevention, 807
725 808
P
Bonneson, J., Pratt, M., Miles, J., & Carlson, P. (2007). Horizontal curve signing handbook. 86, 173–185.
726 Texas Transportation Institute. Texas A&M University System, College Station, Janssens, D., Wets, G., Timmermans, H. J., & Arentze, T. A. (2007). Modelling short-term 809
727 Texas: Texas Department of Transportation. dynamics in activity-travel patterns: Conceptual framework of the Feathers model. 810
728 Boussofiane, A., Dyson, R. G., & Thanassoulis, E. (1991). Applied data envelopment analy- 11th World Conference on Transport Research, Berkeley CA, USA. 811
729 sis. European Journal of Operational Research, 52(1), 1–15. Jiawei, H., & Kamber, M. (2001). Data Mining: Concepts and Techniques. vol. 5.San
D 812
730 Breiman, L., Friedman, J., Stone, C. J., & Olshen, R. A. (1984). Classification and regression Francisco, CA, itd: Morgan Kaufmann. 813
731 trees. CRC press. Joshua, S. C., & Garber, N. J. (1990). Estimating truck accident rate and involvements using 814
732 CARE (2016). Traffic safety basic facts-main figures. Euro Commission Available: https://ec. linear and Poisson regression models. Transportation Planning and Technology, 15(1), 815
E
733 europa.eu/transport/road_safety/sites/roadsafety/files/pdf/statistics/dacota/bfs2016_ 41–58. 816
734 main_figures.pdf. Jovanis, P. P., & Chang, H. -L. (1986). Modeling the relationship of accidents to miles trav- 817
735 Chang, L. -Y. (2005). Analysis of freeway accident frequencies: Negative binomial regres- eled. Transportation Research Record, 1068, 42–51. 818
T
736 sion versus artificial neural network. Safety Science, 43(8), 541–557. Karlaftis, M. G., & Golias, I. (2002). Effects of road geometry and traffic volumes on rural 819
737 Chang, L. -Y., & Chen, W. -C. (2005). Data mining of tree-based models to analyze freeway roadway accident rates. Accident Analysis & Prevention, 34(3), 357–365 5. 820
738 accident frequency. Journal of Safety Research, 36(4), 365–375. Kashani, A. T., & Mohaymany, A. S. (2011). Analysis of the traffic injury severity on two- 821
C
739 Chang, L. -Y., & Chien, J. -T. (2013). Analysis of driver injury severity in truck-involved ac- lane, two-way rural roads based on classification tree models. Safety Science, 49 822
740 cidents using a non-parametric classification tree model. Safety Science, 51(1), 17–22. (10), 1314–1320 12. 823
741 Charnes, A., Cooper, W. W., & Rhodes, E. (1978). Measuring the efficiency of decision mak- Kaye, S. -A., Lewis, I., & Freeman, J. (2018). Comparison of self-report and objective mea- 824
Q10
E
742 ing units. European Journal of Operational Research, 2(6), 429–444. sures of driving behaviour and road safety: A systematic review. Journal of Safety 825
743 Chimba, D., & Kutela, B. (2014). Scanning secondary derived crashes from disabled and Research. 826
744 abandoned vehicle incidents on uninterrupted flow highways. Journal of Safety Kim, D. -G., Lee, Y., Washington, S., & Choi, K. (2007). Modeling crash outcome probabil- 827
R
745 Research, 50, 109–116 9. ities at rural intersections: Application of hierarchical binomial logistic models. 828
746 Clark, L. A. (1992). Tree-based models. Statistical models. Accident Analysis & Prevention, 39(1), 125–134. 829
747 Dadashova, B., Arenas-Ramírez, B., Mira-McWilliams, J., & Aparicio-Izquierdo, F. (2016). Kockelman, K. M., & Kweon, Y. -J. (2002). Driver injury severity: An application of ordered 830
748 Methodological development for selection of significant predictors explaining fatal probit models. Accident Analysis & Prevention, 34(3), 313–321. 831
R
749 road accidents. Accident Analysis & Prevention, 90, 82–94. Kononov, J., Lyon, C., & Allery, B. (2011). Relation of flow, speed, and density of urban free- 832
750 Dissanayake, S., & Lu, J. (2002). Analysis of severity of young driver crashes: Sequential bi- ways to functional form of a safety performance function. Transportation Research 833
751 nary logistic regression modeling. Transportation Research Record: Journal of the Record: Journal of the Transportation Research Board(2236), 11–19. 834
O
752 Transportation Research Board(1784), 108–114. Ladron de Guevara, F., Washington, S., & Oh, J. (2004). Forecasting crashes at the planning 835
753 Dong, C., Clarke, D. B., Yan, X., Khattak, A., & Huang, B. (2014). Multivariate random- level: Simultaneous negative binomial crash model applied in Tucson, Arizona. 836
754 parameters zero-inflated negative binomial regression model: An application to esti- Transportation Research Record: Journal of the Transportation Research Board(1897), 837
755 838
C
mate crash frequencies at intersections. Accident Analysis & Prevention, 70, 320–329. 191–199.
756 Dyson, R. G., & Thanassoulis, E. (1988). Reducing weight flexibility in data envelopment Lee, C., Saccomanno, F., & Hellinga, B. (2002). Analysis of crash precursors on instru- 839
757 analysis. Journal of the Operational Research Society, 39(6), 563–576. mented freeways. Transportation Research Record: Journal of the Transportation 840
758 Eksler, V., & Lassarre, S. (2008). Evolution of road risk disparities at small-scale level: Ex- Research Board(1784), 1–8. 841
N
759 ample of Belgium. Journal of Safety Research, 39(4), 417–427. Lee, S. (2010). Using data envelopment analysis and decision trees for efficiency analysis 842
760 Elvik, R. (1997). Effects on accidents of automatic speed enforcement in Norway. and recommendation of B2C controls. Decision Support Systems, 49(4), 486–497. 843
761 Transportation Research Record: Journal of the Transportation Research Board(1595), Lord, D., Manar, A., & Vizioli, A. (2005). Modeling crash-flow-density and crash-flow-V/C 844
U
762 14–19. ratio relationships for rural and urban freeway segments. Accident Analysis & 845
763 Elvik, R., Christensen, P., & Amundsen, A. (2004). Speed and road accidents. An evaluation Prevention, 37(1), 185–199. 846
764 of the Power Model. TØI report. Vol. 740. Vol. 2004. Lord, D., & Mannering, F. (2010). The statistical analysis of crash-frequency data: A review 847
765 ETSC (2001). Transport Safety Performance Indicators. Brussels: European Transport Safety and assessment of methodological alternatives. Transportation Research Part A: Policy 848
766 Council. and Practice, 44(5), 291–305. 849
767 EU (2016). Traffic safety basic facts-motorways. European Road Safety Observatory Avail- Lord, D., Washington, S. P., & Ivan, J. N. (2005). Poisson, Poisson-gamma and zero-inflated 850
768 able: https://ec.europa.eu/transport/road_safety/sites/roadsafety/files/pdf/statistics/ regression models of motor vehicle crashes: Balancing statistical fit and theory. 851
769 dacota/bfs2016_motorways.pdf. Accident Analysis & Prevention, 37(1), 35–46. 852
770 Frantzeskakis, J. M., & Iordanis, D. I. (1987). Volume-to-capacity ratio and traffic accidents Luoma, J., & Sivak, M. (2007). Characteristics and availability of fatal road-crash databases 853
771 on interurban four-lane highways in Greece. Transportation Research Record(1112). in 20 countries worldwide. Journal of Safety Research, 38(3), 323–327. 854
772 Fu, R., Guo, Y., Yuan, W., Feng, H., & Ma, Y. (2011). The correlation between gradients of Malyshkina, N., & Mannering, F. (2008). Effect of increases in speed limits on severities of 855
773 descending roads and accident rates. Safety Science, 49(3), 416–423 3. injuries in accidents. Transportation Research Record: Journal of the Transportation 856
774 Garber, N., & Ehrhart, A. (2000). Effect of speed, flow, and geometric characteristics on Research Board(2083), 122–127. 857
775 crash frequency for two-lane highways. Transportation Research Record: Journal of Martin, J. -L. (2002). Relationship between crash rate and hourly traffic flow on interur- 858
776 the Transportation Research Board(1717), 76–83. ban motorways. Accident Analysis & Prevention, 34(5), 619–629. 859
860 Miller, J., & Ripley, E. (1996). Layered intrusions of the Duluth complex, Minnesota, USA. Torbic, D. J., et al. (2004). Guidance for implementation of the AASHTO strategic highway 944
861 Developments in Petrology, 15, 257–301. safety plan. Volume 7: A guide for reducing collisions on horizontal curves0309087600.. 945
862 Milton, J. C., & Mannering, F. L. (1996). The relationship between highway geometrics, traffic Wang, B., Hallmark, S., Savolainen, P., & Dong, J. (2017). Crashes and near-crashes on hor- 946
863 related elements, and motor vehicle accidents. izontal curves along rural two-lane highways: Analysis of naturalistic driving data. 947
864 Mohan, D., Bangdiwala, S. I., & Villaveces, A. (2017). Urban street structure and traffic Journal of Safety Research, 63, 163–169. 948
865 safety. Journal of Safety Research, 62, 63–71. Wang, C., Quddus, M. A., & Ison, S. G. (2009). Impact of traffic congestion on road acci- 949
866 Mustakim, F., & Busu, A. (2007). Identifying, prioritizing and treating hazardous location dents: A spatial analysis of the M25 motorway in England. Accident Analysis & 950
867 at Federal Route FT023 (Muar-Segamat). Malaysian Road Conference, 7th, 2007, Prevention, 41(4), 798–808. 951
868 Kuala Lumpur, Malaysia. Wang, H., Li, J., Chen, Q. -Y., & Ni, D. (2011). Logistic modeling of the equilibrium speed– 952
869 Mustakim, F., & Fujita, M. (2011). Development of accident predictive model for rural density relationship. Transportation Research Part A: Policy and Practice, 45(6), 953
870 roadway. World Academy of Science, Engineering and Technology, 58(2011), 126–131. 554–566. 954
871 Noland, R. B., & Oh, L. (2004). The effect of infrastructure and demographic change on Wang, J. -j., & Chen, F. -a. (2012a). Risk evaluation of expressway traffic safety under 955
872 traffic-related fatalities and crashes: a case study of Illinois county-level data. “prior restraint” idea. CICTP 2012: Multimodal transportation systems—Convenient, 956
873 Accident Analysis & Prevention, 36(4), 525–532. safe, cost-effective, efficient (pp. 2169–2180). 957
874 OECD (2002). Safety on Roads: What's the Vision? Paris, France: The Organisation for Eco- Wang, J. -j., & Chen, F. -a. (2012b). Sensitivity analysis of expressway traffic safety risk. 958
875 nomic Co-operation and Development (OECD). CICTP 2012: Multimodal transportation systems—Convenient, safe, cost-effective, efficient 959
876 Ogwueleka, F. N., Misra, S., Ogwueleka, T. C., & Fernandez-Sanz, L. (2014). An artificial (pp. 2219–2229). 960
877 neural network model for road accident prediction: a case study of a developing Wang, Y., & Kockelman, K. M. (2013). A Poisson-lognormal conditional-autoregressive 961
878 962
F
country. Acta Polytechnica Hungarica, 11(5), 177–197. model for multivariate spatial analysis of pedestrian crash counts across neighbor-
879 Olutayo, V., & Eludire, A. (2014). Traffic accident analysis using decision trees and neural hoods. Accident Analysis & Prevention, 60, 71–84. 963
880 networks. International Journal of Information Technology and Computer Science Wegman, F. (2014). Analyzing road design risk factors for run-off-road crashes in the 964
O
881 (IJITCS), 6(2), 22. Netherlands with crash prediction models. Journal of Safety Research, 49, 121. e1–127. 965
882 Pande, A., & Abdel-Aty, M. (2006). Assessment of freeway traffic parameters leading to Wegman, F., et al. (2008). SUNflowerNext: Towards a composite road safety performance 966
883 lane-change related collisions. Accident Analysis & Prevention, 38(5), 936–948. index. Deliverable D6-SWOV. 16.. The Netherlands: Leidschendam. 967
884 Papadimitriou, E., Yannis, G., Bijleveld, F., & Cardoso, J. L. (2013). Exposure data and risk WHO (2004). World Report on Road Traffic Injury Prevention. Switzerland: World Health 968
O
885 indicators for safety performance assessment in Europe. Accident Analysis & Organization Geneva. 969
886 Prevention, Vol. 60, 371–383. WHO (2006). The world health report 2006: working together for health. World Health 970
887 Persaud, B., Retting, R., & Lyon, C. (2000). Guidelines for identification of hazardous high- Organization. 971
R
888 way curves. Transportation Research Record: Journal of the Transportation Research WHO (2015). Global Status Report on Road Safety 2015. Italy: World Health Organization. 972
889 Board(1717), 14–18. Wong, Y. -H., & Beasley, J. (1990). Restricting weight flexibility in data envelopment anal- 973
890 Rahimi, I., Behmanesh, R., & Yusuff, R. M. (2013). A hybrid method for prediction and as- ysis. Journal of the Operational Research Society, 41(9), 829–835. 974
891 sessment efficiency of decision making units: Real case study: Iranian poultry farms. Wu, D. (2009). Supplier selection: A hybrid model using DEA, decision tree and neural 975
P
892 International Journal of Decision Support System Technology (IJDSST), 5(1), 66–83. network. Expert Systems with Applications, 36(5), 9105–9112. 976
893 Rosić, M., Pešić, D., Kukić, D., Antić, B., & Božović, M. (2017). Method for selection of opti- Yasin Çodur, M., & Tortum, A. (2015). An artificial neural network model for highway ac- 977
894 mal road safety composite index with examples from DEA and TOPSIS method. cident prediction: A case study of Erzurum, Turkey. PROMET-Traffic&Transportation, 978
895 Accident Analysis & Prevention, 98, 277–286 1. 27(3), 217–225.
D 979
896 Rumar, K. (1999). Transport safety visions, targets and strategies: beyond 2000. 1st Ye, F., & Lord, D. (2011). Investigation of effects of underreporting crash data on three 980
897 European Transport Safety Lecture. Tech. Rep. Brussels: European Transport Safety commonly used traffic crash severity models: multinomial logit, ordered probit, 981
898 Council. and mixed logit. Transportation Research Record: Journal of the Transportation 982
E
899 SafetyNet, W. (2005). State of the Art Report on Road Safety Performance Indicators. EU In- Research Board(2241), 51–58. 983
900 tegrated Project SafetyNet. Yeo, H., Jang, K., Skabardonis, A., & Kang, S. (2013). Impact of traffic states on freeway 984
901 Samoilenko, S., & Osei-Bryson, K. -M. (2008). Increasing the discriminatory power of DEA crash involvement rates. Accident Analysis & Prevention, 50, 713–723. 985
T
902 in the presence of the sample heterogeneity with cluster analysis and decision trees. Zhou, M., & Sisiopiku, V. (1997). Relationship between volume-to-capacity ratios and 986
903 Expert Systems with Applications, 34(2), 1568–1581. accident rates. Transportation Research Record: Journal of the Transportation Research 987
904 Samoilenko, S., & Osei-Bryson, K. -M. (2013). Using Data Envelopment Analysis (DEA) for Board(1581), 47–52. 988
C
905 monitoring efficiency-based performance of productivity-driven organizations: De-

906 sign and implementation of a decision support system. Omega, 41(1), 131–142.
Syyed Adnan Raheel Shah is currently working as an Assistant Professor at the Depart- 989
907 Schneider IV, W., Zimmerman, K., Van Boxel, D., & Vavilikolanu, S. (2009). Bayesian anal-
ment of Civil Engineering, PIET, Multan. He did his BSc and MSc in Civil Engineering from 990
908
E
ysis of the effect of horizontal curvature on truck crashes using training and valida-
NUST Islamabad. Currently he is joint Ph.D scholar in UET Taxila and Hasselt University, 991
909 tion data sets. Transportation Research Record: Journal of the Transportation Research
Belgium. Syyed Adnan has more than 7 years of work experience in the field of Transpor- 992
910 Board(2096), 41–46.
tation Engineering. He has produced a number of publications from his research work. 993
911
R
Schneider IV, W. H., Savolainen, P. T., & Moore, D. N. (2010). Effects of horizontal curva-
994
912 ture on single-vehicle motorcycle crashes along rural two-lane highways.
913 Transportation Research Record, 2194(1), 91–98. Naveed Ahmad is currently working as an Associate Professor at the Department of Civil 995
914 Seol, H., Choi, J., Park, G., & Park, Y. (2007). A framework for benchmarking service process Engineering, University of Engineering and Technology Taxila. Dr. Naveed was born in 996
R
915 using data envelopment analysis and decision tree. Expert Systems with Applications, Pakistan in 1980. He did his BSc Honors and MSc in Civil Engineering from UET Taxila. 997
916 32(2), 432–440. He holds a Ph.D. degree from the University of Nottingham, UK. Dr. Naveed worked as a 998
917 Sexton, T. R., Silkman, R. H., & Hogan, A. J. (1986). Data envelopment analysis: Critique Research Associate at the University of Nottingham UK, and as a Post-doctoral Research 999
1000
O
918 and extensions. New Directions for Evaluation, 1986(32), 73–105. Fellow on an EPSRC UK funded Joint research project between the University of Notting-
919 Shafaghizadeh, S., & Attari, M. Y. N. (2014). An integrated fuzzy data envelopment analy- ham and Imperial College London. He also worked as a Visiting Scholar with the research 1001
920 sis and data mining for performance assessment of insurance branches. Recent Devel- team at Texas Transportation Institute, Texas A&M University, USA. Dr. Naveed has more 1002
921 opments in Data Envelopment Analysis and its Applications (pp. 64). than 13 years of work experience in the field of Transportation Engineering. He has pro- 1003
C
922 Shah, S. A. R., Ahmad, N., Shen, Y., Pirdavani, A., Basheer, M. A., & Brijs, T. (2018). Road duced a number of publications from his research work. He is also a member of different 1004
923 safety risk assessment: An analysis of transport policy and management for low-, national and international engineering councils/societies and research groups. 1005
924 middle-, and high-income Asian countries. Sustainability, 10(389). 1006
N
925 Shah, S. A. R., Brijs, T., Ahmad, N., Pirdavani, A., Shen, Y., & Basheer, M. A. (2017). Road Yongjun Shen is a professor at the School of Transportation of Southeast University in 1007
926 safety risk evaluation using GIS-based data envelopment analysis—artificial neural China. He received his Ph.D in Transportation Sciences from Transportation Research Insti- 1008
927 networks approach. Applied Sciences, 7(9), 886. tute (IMOB) of Hasselt University in Belgium. His current research interests are traffic 1009
1010
U
928 Shankar, V., & Mannering, F. (1996). An exploratory multinomial logit analysis of single- safety and sustainability, travel behavior modeling and risk analysis, driving simulator re-
929 vehicle motorcycle accident severity. Journal of Safety Research, 27(3), 183–194. search, data mining and decision making. He has participated in a number of EU, Flemish, 1011
930 Shen, Y., Hermans, E., Brijs, T., Wets, G., & Vanhoof, K. (2012). Road safety risk evaluation and Chinese projects, and has published over 60 articles in scientific peer-reviewed 1012
931 and target setting using data envelopment analysis and its extensions. Accident journals, books, and conferences, such as Transportation Research Part C, Accident Analy- 1013
932 Analysis & Prevention, 48, 430–441. sis and Prevention, Expert Systems with Applications, etc. He currently serves as a guest 1014
933 Siddique, R., Aggarwal, P., & Aggarwal, Y. (2011). Prediction of compressive strength of editor of several international journals, such as Transportation Research Part D, and as in- 1015
934 self-compacting concrete containing bottom ash using artificial neural networks. ternational program committee member of several international conference series, such 1016
935 Advances in Engineering Software, 42(10), 780–786. as FLINS, ISKE, HPCC, etc. 1017
936 Sohn, S. Y., & Lee, S. H. (2003). Data fusion, ensemble and clustering to improve the clas- 1018
937 sification accuracy for the severity of road traffic accidents in Korea. Safety Science, 41 Mumtaz Ahmed Kamal is a Professor and Dean at the Department of Civil Engineering, 1019
938 (1), 1–14. University of Engineering and Technology Taxila.. He received his Ph.D in Transportation 1020
939 Sohn, S. Y., & Moon, T. H. (2004). Decision tree based on data envelopment analysis for Engineering from The Queen's University of Belfast, UK. His current research interests 1021
940 effective technology commercialization. Expert Systems with Applications, 26(2), are traffic safety and sustainability, behavior of Unbound Granular Materials in flexible 1022
941 279–284. pavement, Studying the seasonal effects using Falling Weight Deflectometer and the 1023
942 Sze, N. -N., & Wong, S. (2007). Diagnostic analysis of the logistic model for pedestrian in- effects of repeated load on resilient and permanent deformation on A.C under local 1024
943 jury severity in traffic crashes. Accident Analysis & Prevention, 39(6), 1267–1278. environmental conditions and the functional performance of landing Air strips and 1025
1026 decision-making. He has participated in a number of UK and Pakistani projects, and has simulation. He graduated in 1997 from the Limburg University Center (now: Hasselt 1036
1027 published over 75 articles in scientific peer-reviewed journals and conferences. University) as Engineer in Business Informatics. From 1997 until 2002 he was active 1037
1028 at the Limburg University Center as a Ph.D student. He obtained his Ph.D. title in data 1038
1029 Muhammad Aamir Basheer did his BSc from UET and MS in Transportation Science from mining in 2002. Since then, he started working for Hasselt University's Transportation 1039
1030 Hasselt University, Belgium. Currently he is Ph.D scholar in Ghent University, Belgium. Research Institute (IMOB) as an FWO postdoctoral researcher and lecturer with road 1040
1031 Aamir has more than 3 years of work experience in the field of Planning. He has produced safety as his core research area. He published over 80 scientific articles in peer- 1041
1032 a number of publications from his research work. reviewed ISI ranked academic journals. He also serves as a member of the editorial 1042
1033 board of several international scientific journals in the field of road safety and driving 1043
1034 Tom Brijs is full professor in road safety, vice director of IMOB and coordinates scien- simulation, including: Accident Analysis and Prevention, Advances in Transportation 1044
1035 tific research in the areas of road safety, human-centered road design and driving Studies, and Safety. 1045
1046
F
O
O
R
P
D
E
T
C
E
R
R
O
C
N
U

Journal of Safety Research

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Journal of Safety Research

Uploaded by

Copyright:

Available Formats

JSR-01556; No of Pages 15

Journal of Safety Research xxx (xxxx) xxx

Contents lists available at ScienceDirect

Journal of Safety Research

journal homepage: www.elsevier.com/locate/jsr

1Q2 Relationship between road trafﬁc features and accidents: An application

it is in top 10 worst countries among the peer European countries, as 61

To evaluate the road safety condition of roads, a two-stage perfor- 343

Fig. 4. DEA-DT framework for accident risk evaluation.

Rating DMU Rated DMU t2:3

1 E11 E12 E13 … E1n t2:5

t1:2 Data description for motorways of Belgium Limburg section.

t1:3 Road Variable Units N Mean SD Min. Med. Max.

t1:7 Vert_Curve 1-Up, 2-Down, 3-No 36 1.917 0.732 1 2 3

t3:5 1 0.369 3039.22 1541.61 74 105 60.79 91.07 4.51 1

can opt for all road segments. 544

SPEED 1 25.04 0.4844 t5:5

t6:3 Leaf label Leaf Mean Count

t6:4 SPEED b118.86&Vert_Curve b3&FLOW≥ 1197.07&FLOWb1276.10 1 0.66304286 7

Fig. 6. Accident severity risk map for motorway segments.

Fig. 7. Road safety risk prediction through decision tree model.

4.3.4. Vertical curve 612

585 4.3.2. Flow

Fig. 9. Impact analysis of factors on risk-severity.

640 of view is clear. ment budgets. 673

641 5. Conclusions Acknowledgments 674

642 This study focuses on accident analysis of road infrastructure geom-

645 assessment (i.e., a benchmarking instrument of DEA combined with a

905 monitoring efﬁciency-based performance of productivity-driven organizations: De-

You might also like