sustainability
Article
The Impact of Road Geometric Formation on Traffic Crash and
Its Severity Level
Debela Jima 1, *
and Tibor Sipos 2,3
1
2
3
*
Citation: Jima, D.; Sipos, T. The
Impact of Road Geometric Formation
on Traffic Crash and Its Severity
Level. Sustainability 2022, 14, 8475.
https://doi.org/10.3390/
su14148475
Academic Editors: Marco Guerrieri
and Matjaž Šraml
Department of Transportation and Vehicle Engineering, Budapest University of Technology and Economics,
Műegyetem Rakpart 3, 1111 Budapest, Hungary
Department of Transport Technology and Economics, Budapest University of Technology and Economics,
Műegyetem Rakpart 3, 1111 Budapest, Hungary; sipos.tibor@kjk.bme.hu or sipost@kti.hu
KTI—Institute for Transport Sciences, Directorate for Strategic Research and Development,
1119 Budapest, Hungary
Correspondence: Debela.Jima@edu.bme.hu
Abstract: Road infrastructure has an impact on the occurrence of road traffic crashes. The aim of this
study was to analyze the impact of road geometric formation on road traffic crashes. Based on the
nature, convenience, and availability of data, the study used Budapest city road traffic crash data from
2017 to 2021. For organizing, analysis, and modeling, the study used Microsoft-Excel, the Statistical
Package for Social Science, and Quantum Geographic Information System. Relative frequency
distribution, Multinomial Logistic Regression, Multilayer Perceptron Artificial Neural Network, and
Severity Index were used for the analysis. Both inferential and descriptive statistics are used to
describe and summarize the study outcome. Multicollinearity tests, p-value, overdispersion, percent
of incorrect error, and other statistical model testes were undertaken to analyze the significance of the
data and variable for modeling and analysis. A large number of crashes were observed in straight
and one-lane road geometric formationsr890. However, the severity level was high at the horizontal
curve and in all three lanes of the road. The regression model indicated that light conditions, collision
type, road geometry, and speed had a significant effect on traffic accidents at a p-value of 0.05. A
collision between the vehicle (rear end collision), and a vehicle with a pedestrian was the probable
cause of the crash. The Multilayer Perceptron Artificial Neural Network indicated that horizontally
curved geometry has a positive and strong relationship with road traffic fatalities. The primary
reasons for the occurrences of a road traffic crash at an intersection, horizontal curve, and straight
road geometric formation were the improper use of road traffic signs, road pavement condition, and
stopping sight distance problems, respectively. The hourly distribution showed that from 16:01 to
17:00 time interval was a peak hour for the occurrences of road traffic crashes. Whereas, driver plays
vital role and responsible body for the occurrences of crashes at all geometric formations.
Received: 13 May 2022
Accepted: 7 July 2022
Published: 11 July 2022
Keywords: multilayer perceptron artificial neural network; multinomial logistic regression; quantum
geographic information system; road traffic crash; severity index
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affiliations.
Copyright: © 2022 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
1. Introduction
Unsafe and insufficient road infrastructure is a fundamental issue for the occurrence of
road traffic crashes (RTC) and their outcome. RTC occurrences are significantly influenced
by road geometric formation [1]. It can be fragmented into alignment, profile, and crosssection. Mostly, road width, cross slope, road margins, traffic separators, and curbs can be
considered basic physical elements [2,3]. The intention of geometric design is to optimize
efficiency and safety so that it minimizes cost and environmental damage.
To analyze the impact of road geometric formation on traffic crashes and their severity
levels, the study used recent 5-year Budapest city road traffic crash data that was collected
from Hungary’s transport authority from 2017 to 2021. The study area was selected due
to different factors such as the nature of the data, availability, and convenience of data. In
Sustainability 2022, 14, 8475. https://doi.org/10.3390/su14148475
https://www.mdpi.com/journal/sustainability
Sustainability 2022, 14, 8475
2 of 25
Budapest city, there was a high concentration of traffic crashes that were recorded yearly.
In the past 5 years, around 17,006 road traffic crashes have been registered. The crash’s
outcome has been classified as fatality (216), serious injury (3999), and minor injury (12,791).
Different studies showed that road infrastructure, mostly geometry formation, had its
own impacts on road traffic crashes. Pembuain et al. (2018) indicated that the elements in
road infrastructure formation had a significant effect on the risk of road traffic accidents [4].
The report in Australia showed that the road is a causation factor in about 30% of all
crashes [5]. A road defect directly triggers a crash, where some element of the road environment misleads a road user and thereby creates human errors [6]. The study showed that
two-lane rural highways reduce crash rates by 44% versus high crash-rate infrastructure, at
the 99% confidence level [7]. One of the important and cogent measures for reducing road
crash fatalities is continuously improving and maintaining the good shape and condition
of our roads [8]. Road safety can be improved by implementing principles of road safety
infrastructure management (RIS) [9]. It showed that road geometry elements can mislead
road users. Two-lane roads highly contribute to crashes; in this case, it is important to
examine the impact of other lane formations on RTC and its severity level.
A study in Texas indicated that severe crashes are likely to occur on horizontal
curves with higher degrees of curvature compared to curves with smaller degrees of
curvature [10–12]. Sarbaz Othman et al. (2009) stated that large-radius right-turn curves
were more dangerous than left curves during lane-changing maneuvers. However, sharper
curves are more dangerous in both left and right curves [13]. Overtaking on right curves
was sensitive to the radius and the interaction of the radius with road conditions, while the
left curves were more sensitive to super-elevation [14]. Even though the study shows that
horizontal curves are a cause of severe road traffic crashes, it is better to analyze the overall
impact of curved roads on RTC compared to other road geometric formations.
On straight roads, speed and distance were influenced by road traffic accidents. The
longest distance offered the highest risk of fatal injuries [15]. Willy et al. (2020) discovered a strong relationship between side freedom and accident number [16]. MBESSA
Michel et al. (2020) stated that the reduction in road geometric formation has a significant
impact on crash occurrence [17]. In Singapore, road crashes at intersections contribute
around 35% of the reported accidents and show that vehicle type, road type, collision
type, driver’s characteristics, and time of day are important determinants of the severity of
crashes at intersections [18]. The study in the U.S. showed that the relative ratio analysis
showed that intersection-related crashes were almost 335 times as likely to have “turned
with an obstructed view” as the critical reason for non-intersection-related crashes [19].
So, road geometric formations, such as horizontal curves, side freedom, intersections, and
straight roads have their own impact on the occurrences of road traffic crashes. It was
better to visualize the interaction between stated variables and the number of lanes with
intersection type.
Abbasi et al. (2022) stated that developing artificial lighting at intersections and LEDraised pavement markers on two-lane rural roads could lead to enhanced road safety under
dark LCs [20]. Mehdi Hosseinpour et al. (2013) study results of REGOPM on crash severity
showed that horizontal curvature, paved shoulder width, terrain type, and side friction
were associated with more severe crashes, whereas land use, access points, and the presence
of a median reduced the probability of severe crashes [21,22]. In this case, further analysis
was required to define the effect of natural light conditions on RTC.
Different research on road geometric formation and its impact on road traffic crashes
showed different attributes and their level of contribution. Most of the studies focused
on a microscopic level, such as selected road segments, rural road networks, intersections,
specific numbers of lanes, small road networks, etc. According to my knowledge, there
was a research gap on the impact of road geometric formation on road traffic crashes and
their severity levels at a macroscopic level, such as at the country or city level. Budapest
City was selected due to the fact that the area was urban, with highly networked roads,
large territory, and has a high concentration of vehicles and a dense population. In fact, the
Sustainability 2022, 14, 8475
3 of 25
number of road traffic accidents recorded was also high, and the nature of the data and its
accessibility encourages the choice of the city. Moreover, as per study review in Budapest
city, the impact of road geometric formation on road traffic crashes at the macroscopic
level and their severity rate was not taken into consideration. In spite of that, further
investigation was needed to direct the impacts of road geometric formation on road traffic
crashes and their severity levels.
The main objective of this study was to analyze the impact of road geometric formation
on road traffic crashes and their severity levels. For further analysis, the study emphasizes
the road’s geometric formation (straight, intersection, horizontal curvature, number of
lanes, etc.) and number of lanes. In addition to that, this study tried to highlight the
probable causes of road traffic crashes and the relationship between variables. The study
used relative frequency distribution for statistical analysis, multinomial logistic regression
to analyze the main determinant factor of a road traffic crash, and multilayer perceptron
artificial neural network (MLP-ANN) to demonstrate the influences and interaction of road
geometric formation and the number of lanes on a road traffic crash and its outcome. In
addition to that, it used the severity index (SI) to show the severity level of road geometric
formation and the number of lanes that lead to a road traffic crash and its outcome. For
explanation, description, and summarizing the output, the study used both inferential and
descriptive statistics.
2. Material and Method
2.1. Data Type, Source and Method of Collection
This study used road traffic crash data collected by the Hungary government’s Budapest Transportation Authorities as a secondary data source. As a result, Budapest city
road traffic crash data was considered as an input for further analysis of the impacts of
unsafe and insufficient road geometric formation on traffic crashes. The study area was
selected because of the nature of the data, availability, and convenience of the data. In
addition to that, there were high and dense traffic crashes registered yearly. To achieve
significant results and fulfill the minimum requirement, the study used 5-year data from
2017 to 2021 [20].
For data management and analysis, the study used tools such as Ms. Excel for data
organization, a statistical package for social science (SPSS-20) for data analysis and modeling, and Q-GIS for the analysis of location-related information, etc. Each variable and
parameter were coded according to data type and priority.
2.2. Variable Definition
Depending on the objective of the study and the type of data collected from the
authorities, these studies consider variables as dependent and independent to facilitate
the analysis. Accordingly, it considers road crashes (outcome) as dependent variables.
Whereas the independent variable was described as hourly distribution, collision type,
light condition, causes of a crash, geometric formation, pavement surface, number of lanes,
speed, weather condition, alcohol consumption, responsible body, etc.
2.3. Method of Analysis
The study used relative frequency distribution to further investigate the occurrences
and rates of road traffic crashes [23,24]. This study also used multinomial logistic (MNL)
regression to analyze the determinant factors of road traffic crashes. To analyze the severity
level of road geometric formation and the number of lanes that cause road traffic crashes
and their outcomes, the study used the Severity Index (SI). Furthermore, the Multilayer
Perceptron Artificial Neural Network (MLP-ANN) was used to show the impacts of road
geometric formation and their relationship with the severity level of road traffic crash
outcomes.
This study also used the Quantum Geographic Information System (Q-GIS) to enable
the location of road traffic crashes and analyze spatial information [25]. It also used
Sustainability 2022, 14, 8475
4 of 25
combined geographical, statistical, and mapping data in the study [26,27]. That was
used for mapping road traffic injuries, accident analysis, and the determination of hot
spots [28,29].
For a detailed explanation and interpretation of the output, the study used inferential
statistics. It also used descriptive statistics to summarize the characteristics of a sample or
data set, such as frequency, since it helps us to understand the features of a specific data set
by giving short summaries of the sample and measures of the data [30,31].
2.3.1. Severity Index
To analyze the severity level of road geometric formation that causes road traffic
crashes and their outcomes, the study used the Severity Index (SI). Empirically, the crash
severity index is expressed as shown in Equation (1) below [32].
Seveity Index (SI ) =
Number o f Injuries
Number o f Death
or
Total Number o f Crash
Total Number o f Crash
(1)
2.3.2. Multinomial Logistic Regression
Based on the nature of the data, this study used Multinomial Logistic (MNL) Regression to analyze the determinant factor of road traffic crashes. It was an appealing
statistical approach in modeling the severity of road traffic crashes because it allows for
more than two categories of the dependent or outcome variable and does not require the
assumption of normality, linearity, or homoscedasticity [23,24]. It is also used to investigate
multi-vehicle collisions in different forms and is appropriate for both non-interstate and
interstate crashes involved in [33,34].
The model assumes that there is a series of observations (dependent variable) Ai for
i = 1, 2 . . . n. Along with each observation, Ai , there is a set of observed values X1 , . . . , Xn
of explanatory variables. The output Ai is categorically distributed based on the outcome of
the crash. So, this study categorized road traffic crashes and their outcomes as dependent
variables and the others as independent variables. The outcomes of the crash were fatalities,
serious injuries, and slight injuries [31].
Ai \ Xi . . . . . . . . . . . . . . . . . . ..Xn , f or i = 1, 2, 3 . . . . . . .n
(2)
Multinomial Logistic Regression is often written in terms of a latent variable model as
stated below.
A1i ∗ = β 1∗ Xi + ε 1
A2i ∗ = β 2∗ Xi + ε 2
...........................
Ain∗ = β n∗ Xi + ε n
where
ε ∼ N 0, ∑
Based on the above relationships, the model that helps to predict road traffic accident
can be defined as a predicting variable A.
A = β + β 1 X1 + β 2 X2 + . . . . . . . . . . . . . . . . . . . . . . . . . + β n X n + ε
(3)
where: A = Dependent (predicted) variable; β = Constant; βi = Intercepts; for i = 1, 2, 3 . . .
. . . . . . n; ε = Error term; X = independent variables.
2.3.3. Multilayer Perceptron Artificial Neural Network (MLP-ANN)
The Multilayer Perceptron Artificial Neural Network (MLP-ANN) is a type of artificial neural network that is used to analyze and model complex patterns and prediction
problems [21,22]. It is also used to show the impacts of road geometric formation and its
Sustainability 2022, 14, 8475
5 of 25
relationship with the severity level of road traffic crash outcomes. It consists of three types
of layers: the input layer, output layer, and hidden layer. The input layer receives the input
signal (data) to be processed [35]. It was applied in a multilayer feed-forward network to
transmit information [36]. It is used to determine its suitability for traffic accident prediction
and to analyze the increasing amounts of traffic accident data and road traffic accident
causes using machine learning [37–39].
2.3.4. Multicollinearity Test
A multicollinearity test was conducted using the Pearson Correlation Coefficient
and, showed that the relationship between most of the variables is not significant; with a
correlation coefficient less than 0.8 indicating the variables can be used for further analysis.
Subjected to information from analysis, a relatively strong relationship was observed
between light conditions and hourly distribution (0.289), alcohol consumption, and collision
type (0.245). However, the ranges are still within the traditional tolerable limit and the
variables can be used for analysis. For more information, see Table A6.
2.3.5. Latent Variable Dispersion
A latent variable is a variable that cannot be observed. The presence of latent variables,
however, can be detected by their effects on variables that are observable. So, checking
overdispersion of both dependent and independent variables is used to select the proper
model to analyze the determinant factor of road traffic crashes. The mean and variances
of variables are used to define overdispersion. As shown in Table 1, below, the mean of
dependent (crash outcome) is greater than the variance. Not only that, more than 80% of
independent variables in this study indicated that their mean is greater than their variances.
Table 1. Mean and Variance of Variables.
Variable
Number
Valid
Missed
Mean
Variances
Dependent
Crash Outcome
17,006
0
2.74
0.218
Independent
Light Condition
Collision Type
Road Geometry
Reason
Pavement Surface
Weather Condition
Alcohol Consumption
Speed Limit
17,006
17,006
17,006
17,006
17,006
17,006
17,006
17,006
0
0
0
0
0
0
0
0
1.27
3.94
2.35
4.62
1.06
2.66
5.05
2.11
0.233
9.204
0.900
9.821
0.119
0.821
0.512
0.218
Table 1 shown above indicates that the rate of overdispersion is low. In addition to
that, the nature of the data also defines model selection. Based on the above evidence, a
multinomial logistic regression model was selected to analyze the determinant factor of
road traffic crashes.
2.3.6. Data Processing Flow Scheme
The general flow scheme of the research methodology for data processing was summarized below (Scheme 1).
Sustainability
2022,REVIEW
14, 8475
ability 2022,
14, x FOR PEER
6 of 26
6 of 25
Scheme 1. DataScheme
Collection,
Management
and
Analysis Process.
1. Data
Collection,
Management
and Analysis Process.
3. Result and Discussion
3. Result and Discussion
of the study
attempt
identify
the analysis
output of
thediscuss
analysis and discuss
These sectionsThese
of thesections
study attempt
to identify
theto
output
of the
and
the
findings.
It
contains
the
road
traffic
crash
distribution
and
its
frequency
the findings. It contains the road traffic crash distribution and its frequency in terms of in terms of
independent
and cross-tabulation
betweenand
dependent
and independent
variables
independent variables,
andvariables,
cross-tabulation
between dependent
independent
variato
signify
the
impact
of
other
variables
on
road
traffic
crashes
and
their
outcome.
Moreover,
bles to signify the impact of other variables on road traffic crashes and their outcome.
it cross-tabulates
independent
variables
against
each
other.
In general,
Moreover, it cross-tabulates
thethe
independent
variables
against
each
other.
In general,
thisthis study tried
bring
reliable
findings
their
implications
define
impact
of road geometric
study tried to to
bring
reliable
findings
andand
their
implications
thatthat
define
the the
impact
of road
formation
traffic
crashes
andtheir
theirseverity
severitylevel.
level. For
For more
more information,
geometric formation
on on
traffic
crashes
and
information, please see the
study
findings
discussed
below:
please see the study findings discussed below:
3.1. Road Network and Traffic Crash Distribution
3.1. Road Network and Traffic Crash Distribution
Figure 1 shown below indicates that since 2017, road traffic crashes in the past 5 years
Figure 1 shown below indicates that since 2017, road traffic crashes in the past 5 years
have been evenly distributed on the road network. It indicated that a high concentration
have been evenly distributed on the road network. It indicated that a high concentration
of road traffic crashes was recorded in the downtown (center) of the city. To locate the
of road traffic crashes was recorded in the downtown (center) of the city. To locate the
concentration of road traffic crashes, the study used Q-GIS.
concentration of road traffic crashes, the study used Q-GIS.
2022, 14, xSustainability
FOR PEER REVIEW
2022, 14, 8475
7 of 26
Sustainability 2022, 14, x FOR PEER REVIEW
7 of 25
7 of 26
Figure
1. Road
Road
Traffic Crash
Crash
Distribution
of
study
area
Figure 1. Road Traffic
Crash
Distribution
of the
study area
from
2017–2021.
Figure
1.
Traffic
Distribution
of the
the
study
areafrom
from2017–2021.
2017–2021.
3.2.
Traffic
Crash
Distribution
and District
DistrictLevel
LevelofofConcentration
Concentration
3.2. Road
Road
TrafficDistribution
Crash Districtand
Distribution
and
3.2. Road Traffic Crash
District
District Level
of Concentration
The study site has 23 districts. The study tried to rate the level of a road
road traffic
traffic crash
crash in
The study site has 23 districts. The study tried to rate the level of a road traffic crash
in percentage
basedononitsitsconcentration.
concentration. As
As a result,
into
three;
percentage
based
result, the
thestudy
studygrouped
groupedthe
thearea
area
into
three;
in percentage based
on 3–6%,
its concentration.
As
a result,
the study
the areaThe
into
three;
0–3%,
and
intermediate,
and
low,
and
0–3%,
3–6%,
and 6–9%
6–9% as
as high,
high,
intermediate,
andgrouped
low,respectively.
respectively.
Themaximum
maximum
and
0–3%, 3–6%, andminimum
6–9% as road
high,traffic
intermediate,
and low,were
respectively.
The
maximum
and
crash
XIV
with
8.5%
and
minimum
road
traffic
crashconcentrations
concentrations
wereobserved
observedin
inDistrict
District
XIV
with
8.5%
and
minimum road traffic
crash
concentrations
were
observed
in District
XIV2 2below
with
8.5%
and
1.9%,
respectively.
information,
see
and
XXIII with
1.9%,
respectively. For
Formore
more
information,
seeFigure
Figure
below
andTable
TableA1.
A1.
XXIII with 1.9%, respectively. For more information, see Figure 2 below and Table A1.
Figure2.
2. (a,b)
(a,b) Road
Road Traffic Crash District
Figure
District Distribution
Distributionand
andLevel
LevelofofConcentration
ConcentrationbybyPercent.
Percent.
Based
on the
theabove
aboveFigure
Figure
color
indicates
a highly
concentrated
road
Based on
2b,2b,
thethe
redred
color
indicates
a highly
concentrated
road traftraffic
crash
was
observed
in districts
XIII,
and
XIV.
Relatively,
greenand
fic crash
thatthat
was
observed
in districts
X, X,
XI,XI,
XIII,
and
XIV.
Relatively,
thethe
greenand
Figure 2. (a,b) Roadorange-colored
Traffic Crash District
Distribution
and
Level
of
Concentration
by
Percent.
districts had a low and intermediate concentration of road traffic crashes
Based on the above Figure 2b, the red color indicates a highly concentrated road traffic crash that was observed in districts X, XI, XIII, and XIV. Relatively, the green- and
Sustainability 2022, 14, 8475
8 of 25
orange-colored districts had a low and intermediate concentration of road traffic crashes
and their outcomes. As a result, further investigation and remedial action were needed in those
stated districts by the stakeholders to minimize road traffic crash severity in the study area.
3.3. Road Geometry Formation and Frequency of Road Traffic Crash
This part of the study deals with the frequency of road traffic crashes in line with the
geometric form of the road. Table 2 shown below indicates how road traffic crashes and
their outcomes vary with road geometric formation.
Table 2. Road Geometry Formation and Traffic Crash Frequency.
Intersection
Horizontal Curve
Straight Road
Others
Total
Frequency
Percent
5319
726
10,825
136
17,006
31.3
4.3
63.7
0.8
100.0
According to Table 2, approximately 63.7% and 31.3% of road traffic accidents occurred on straight and intersection sections of the road network, respectively. Even if
a high number of road traffic crashes are observed on the straight road segments, the
intersection part of the road also contributes to significant road traffic crashes. As a result,
the stakeholders must identify causes and problems on the straight part of the road to
minimize road traffic crashes in the study area. Crashes are not only high at the straight
part of the road, in comparison with areal coverage, the rate of crashes at intersections is
relatively high. Further investigation was also needed at the intersection part of the road.
3.4. Road Geometry Formation and Road Traffic Crash Outcome
As shown in Table 3 below, a high number of fatalities and injuries were registered
on the straight part of the road segment. A relatively significant number of fatalities and
injuries are also registered at intersections. Even though the straight section of the road
had a high number of accidents, in terms of areal coverage, the intersection part of the road
also played a significant role in the occurrences of road traffic accidents. As a result, to
minimize the number of deaths, further investigation was needed on both straight and
intersection parts of the road.
Table 3. Cross-tabulation of Crash Outcome and Road Geometric Formation.
Crash
Outcome
Slight
Injuries
Serious
Injuries
Fatality
Intersection
Horizontal
Curve
Straight
Road
Others
Total
4192
529
7989
81
12,791
1081
177
2693
48
3999
46
20
143
7
216
3.5. Road Geometric Formation and Severity Level of Traffic Crash
Distinct from accident frequency, the crash severity index provides the severity of each
crash outcome registered during a specific time. Based on Equation (1) above, this study
tried to analyze the severity level of the road geometric formation based on the number of
deaths and injuries.
Table 4 shown above indicates that the number of road traffic crash outcomes varies
in different parts of the road section. A high level of severity in terms of death, serious
and slight injuries is observed at the horizontal curve, straight and intersection parts of the
road, respectively. Even if the number of deaths is high on the straight part of the road,
the level of severity in terms of deaths cannot indicate this part of the road. As a result,
Sustainability 2022, 14, 8475
9 of 25
in the specified study area, horizontally curved road geometry was the most severe. In
order to reduce the severity level of the road in the study area, additional research into the
horizontal curve of the road network was required.
Table 4. Road Geometric Formation and Its Severity Level.
Intersection
Horizontal
Curve
Straight
Road
Others
SI @ I
SI @ HC
SI @ SR
4192
529
7989
81
0.788
0.729
0.738
1081
177
2693
48
0.203
0.244
0.249
20
143
7
0.009
0.028
0.013
Slight
Injuries
Serious
Injuries
Fatality
46
Where; SI—Severity Index, I—Intersection, HC—Horizontal Curve, SR—Straight Road.
3.6. Multinomial Logistic (MNL) Regression
On the basis of the nature of the data, this study categorized road traffic crashes and
their outcomes as dependent and independent variables. The outcomes of the crash were
fatalities, serious injuries, and slight injuries.
The result of the analysis indicated that there was a relationship between road traffic
accidents and their potential determinants. The results of the cause-effect analysis of the
variables listed in Table 5 revealed that fatality is highly related to light conditions, collision
type, alcohol consumption, and speed limit. Meanwhile, serious injuries are highly related
to collision type, road geometric formation, and the reason for the occurrence of road traffic
crashes. It considers slight injuries as the reference category. So, light condition, collision
type, alcohol consumption, road geometry formation, speed limit, and reason for road
traffic crashes were determinant factors and had a significant effect on the occurrences of
road traffic accidents at a p-value of 0.05.
Table 5. Determinant Factor of Road Traffic Crash and Its Outcome.
Parameter Estimates
Crash Outcome
β
Std. Error
Wald
df
Sig.
Exp(β)
95% Confidence Interval for
Exp(β)
Lower Bound
Upper Bound
Fatality
Intercept
WC
LC
CT
RG
AC
R
PS
VL
−14.340
0.169
0.505
0.118
−0.038
1.216
0.064
0.262
0.727
0.941
0.076
0.125
0.028
0.085
0.176
0.026
0.139
0.090
232.268
4.917
16.464
17.357
0.202
47.942
6.006
3.559
65.023
1
1
1
1
1
1
1
1
1
0.000
0.027
0.000
0.000
0.653
0.000
0.014
0.059
0.000
1.184
1.657
1.125
0.963
3.374
1.066
1.299
2.069
1.020
1.298
1.064
0.816
2.391
1.013
0.990
1.734
1.375
2.115
1.189
1.136
4.760
1.122
1.705
2.469
Serious
Injuries
Intercept
WC
LC
CT
RG
AC
R
PS
VL
−2.367
0.008
0.126
0.039
0.104
0.031
0.058
0.152
0.013
0.187
0.020
0.038
0.007
0.020
0.027
0.006
0.049
0.040
159.628
0.147
11.251
35.082
25.784
1.251
85.482
9.831
0.098
1
1
1
1
1
1
1
1
1
0.000
0.701
0.001
0.000
0.000
0.263
0.000
0.002
0.754
1.008
1.135
1.039
1.109
1.031
1.060
1.164
1.013
0.969
1.054
1.026
1.066
0.977
1.047
1.059
0.936
1.048
1.222
1.053
1.155
1.088
1.073
1.281
1.096
a. The reference category is: Slight Injuries.
Where; WC—Weather Condition, LC—Light Condition, CT—Collision Type, RG—Road Geometry, AC—Alcohol
Consumption, R—Reason, PS—Pavement Surface, VL—Speed Limit, β—Parameter estimate (Coefficient).
Sustainability 2022, 14, 8475
10 of 25
As a result, stakeholders must concentrate on those factors in order to reduce road
traffic accidents. On the basis of the above findings, the model that helps to determine
the significant level of road traffic accidents can be expressed using the following traffic
accident (A) predicting equation.
A = β + β 1 WC + β 2 LC + β 3 CT + β 4 RG + β 5 AC + β 6 R + β 7 PS + β 8 VL + ε(error )
3.7. Collision Type and Road Traffic Crash Outcome
A collision in transportation happens between a vehicle and an object, a vehicle and a
pedestrian, etc., and that plays a significant role in the occurrences of a road traffic crash.
As shown in Table 6 below, collisions of vehicles with pedestrians result in a high number
of road traffic fatalities. Collisions with vehicles, particularly rear-end collisions, play a
significant role. Not only were there fatalities, but there were also numerous injuries as a
result of rear-end collisions, and vehicle collisions with pedestrians.
Table 6. Cross Tabulation between Collusion Type and Road Traffic Crash Outcome.
Crash Outcome
Collusion
Types
Collision with Animals
Collision with
Pedestrian
Pileup Collision
Collision with Object
Head on Collision
Side Swipe Collision
Side Impact Collision
Rear End Collision
Total
Fatality
Serious
Injuries
Slight
Injuries
Total
0
5
8
13
122
1394
2574
4090
6
7
27
1
4
49
177
114
427
12
302
1568
1456
291
1468
46
1164
5784
1639
412
1922
59
1470
7401
216
3999
12,791
17,006
As a result, a high number of road traffic deaths and injuries were registered due to
collisions happening between vehicles and pedestrians, vehicles and vehicles (rear-end
collision). Further investigation was needed to minimize pedestrian deaths and injuries.
3.8. Artificial Neural Network (ANN): Multilayer Perceptron (MLP)
In this study, MLP-ANN was used to analyze and show the relationship between
dependent and independent variables. To extract the output, the study used Statistical Package for Social Science (SPSS-20). It provides specific information on which variables have a
significant impact on the occurrence and outcome of road traffic accidents. Figure 3 shown
below indicates the impact of road geometric formation and the number of lanes on road
traffic crashes and their outcome.
The model attempted to understand the relationship between the training data and
be evaluated on the test data. In this case, 70% of the data is used for training and 30%
for testing. The output of the model summary between the input of road geometry and
number of lanes and the output of road traffic accidents indicated that the percent of
incorrect predictions for training was 25%. Table 7 below indicates that in both cases, the
incorrect prediction for testing was less than 25%. So, the model was a good fit with a
training and testing error of 25%. This shows that the model prediction level was correct
and accurate above 75% [40].
Sustainability
Sustainability2022,
2022,14,
14,8475
x FOR PEER REVIEW
11 11
ofof2625
Figure 3. Road Geometric Formation and Number of Lane Impact on Road Traffic Crash Outcome.
Figure 3. Road Geometric Formation and Number of Lane Impact on Road Traffic Crash Outcome.
Where; RG—Road Geometric Formation, RG-1—Intersection, RG-2—Horizontal Curve, RG-3—
Where; RG—Road Geometric Formation, RG-1—Intersection, RG-2—Horizontal Curve, RG-3—
Straight Road, RG-5—Other (Unknown). NL—Number of Lane, NL-1—One Lane, NL-2—Two
Straight
Road, RG-5—Other
(Unknown).
NL—Number
Lane, NL-1—One
Lane,
NL-2—TwoOutLane,
Lane, NL-3—Three
Lane, and
Above, and
NL-4—Otherof(Unknown).
Whereas:
CO-1—Crash
NL-3—Three
Lane,
and
Above,
and
NL-4—Other
(Unknown).
Whereas:
CO-1—Crash
Outcome
come (Fatality), CO-2—Crash Outcome (Serious Injuries), and CO-3—Crash Outcome (Slight Inju(Fatality),
CO-2—Crash Outcome (Serious Injuries), and CO-3—Crash Outcome (Slight Injuries).
ries).
Table The
7. Model
Summary
of Error
Computations
the Training
and Testing
Samples.data and
model
attempted
to understand
theBoth
relationship
between
the training
be evaluated on the test data. In this case, 70% of the data is used for training and 30% for
Road Geometry
Number of Lane
testing. The output of the model summary between the input of road geometry and numCrossof
Entropy
Error accidents indicated
7355.623
7287.471
ber Training
of lanes and the output
road traffic
that the percent
of incorrect
Percent
Incorrect
Predictions
25.0%
25.0%
predictions for training was 25%. Table 7 below indicates that in both cases, the
incorrect
Cross
Entropy
ErrorSo, the model 2993.791
prediction for testing was
less
than 25%.
was a good fit with a3103.688
training and
Testing
Percent
Incorrect
Predictions
24.3%
24.2%
testing error of 25%.
This shows
that
the model prediction
level was correct and
accurate
above 75% [40].
Dependent Variable: Crash Outcome
Table 7. Model Summary of Error Computations Both the Training and Testing Samples.
The blue and gray lines in Figure 3 show the positive and negative bond between the
Roadweight
Geometry
of Lane In
dependent and independent variables, whose synaptic
is >0 andNumber
<0, respectively.
this diagram, road geometry,
and
the
number
of
lanes
were
input
(independent
variables),
Cross Entropy Error
7355.623
7287.471
Training
and crash outcome
(road Incorrect
traffic accident)
was output (dependent
variable). 25.0%
Percent
Predictions
25.0%
The intersectionCross
part of
the road
has a strong and
positive contribution
to the ocEntropy
Error
2993.791
3103.688
Testingof road traffic crashes that cause fatalities, whereas the horizontal curve of the
currences
Percent Incorrect Predictions
24.3%
24.2%
road network has a strongDependent
and positive
impactCrash
on the
occurrence of slight injuries. The
Variable:
Outcome
straight section of the road has a positive but relatively minor impact on the occurrence of
trafficThe
accidents.
blue and gray lines in Figure 3 show the positive and negative bond between the
In
addition
to that, one-lane
and two-lane
roads hadweight
a strong
positive
contribution
dependent
and independent
variables,
whose synaptic
is >and
0 and
< 0, respectively.
to
of road
traffic and
injuries.
Simultaneously,
three-lane
and above-average
Inthe
thisoccurrence
diagram, road
geometry,
the number
of lanes were
input (independent
variaroads
have
a
significant
positive
and
strong
impact
on
the
occurrences
of minor injuries.
bles), and crash outcome (road traffic accident) was output (dependent variable).
It alsoThe
hasintersection
a relativelypart
positive
stronger
oncontribution
the occurrences
death.
of theand
roadslightly
has a strong
andimpact
positive
to theofoccurSo,
this
indicates
that
road
geometric
formation
has
an
impact
on
the
occurrence
of
road
rences of road traffic crashes that cause fatalities, whereas the horizontal curve of the road
traffic
injuries.
network has a strong and positive impact on the occurrence of slight injuries. The straight
section of the road has a positive but relatively minor impact on the occurrence of traffic
3.9. Road Traffic Crash Level of Severity and Road Geometric Formation
accidents.
The outcome of the crash analysis indicated that a highly severe accident was registered on a straight road. Based on Table 8 shown below, from total road traffic crash
Sustainability 2022, 14, 8475
12 of 25
outcomes, a high number of deaths and injuries were registered on the straight road sections. Relatively, in line with coverage, the level of accidents and injuries was also high at
the intersection part of the road.
Table 8. Cross Tabulation of Road Traffic Crash Outcome and Road Geometric Formation.
Intersection
Horizontal
Curve
Straight
Road
Others
Total
Fatality
Serious Injuries
Slight Injuries
46
1081
4192
20
177
529
143
2693
7989
7
48
81
216
3999
12,791
Total
5319
726
10,825
136
17,006
As a result, further investigation was needed to minimize the number of deaths and
injuries that happen at straight and intersection parts of the road network in the study area.
3.10. Road Geometric Formation and Related Causes of Road Traffic Crash
As shown in Table A3, a high number of road traffic crashes was registered due to the
improper use of traffic control devices such as traffic signs, signals, and marks. The causes
and primary reason for the occurrences of a road traffic crash at an intersection, horizontal
curve, and straight part of the road segment was improper use of road sign, road pavement
condition, and stopping sight distance problem, respectively. So, improper use of road
traffic control devices has a significant impact on the occurrences of a road traffic crash that
causes enormous loss of life and physical damage.
3.11. Road Geometric Formation and Traffic Collision Type
Road traffic collisions are a probable cause of road traffic crashes and accidents in the
transportation system. Table 9 clearly defines the relationship between road geometric
formation and road traffic collisions. Based on statistical data collected from secondary
sources, collisions between vehicles(rear-end collisions) highly happen in all road geometric
formations of the road network. In addition to that, the collision of vehicles with pedestrians
plays a great role in the occurrence of road traffic crashes and accidents. Mostly, road traffic
crashes happen on straight road segments that are the result of rear-end collisions.
Table 9. Cross Tabulation of Collision Type and Road Geometric Formation.
Intersection
Horizontal
Curve
Straight
Road
Others
Total
Collision with Animals
Collision with Pedestrian
Pileup Collision
Collision with Object
Head on Collision
Side Swipe Collision
Side Impact Collision
Rear End Collision
1
668
112
21
925
6
1011
2575
2
81
41
52
141
4
18
387
10
3258
1485
332
850
49
438
4403
0
83
1
7
6
0
3
36
13
4090
1639
412
1922
59
1470
7401
Total
5319
726
10,825
136
17,006
As a result, a high number of road traffic crashes were registered due to collisions
happening between vehicles, and vehicles with pedestrians on a stated road geometric
formation. Mostly, vehicle collisions, such as rear-end collisions, play a significant role in
the occurrences of road traffic crashes on straight and intersection segments of the road.
Sustainability 2022, 14, 8475
13 of 25
3.12. Road Geometric Formation and Traffic Crash Hourly Distribution
Table A2 indicated that in 1-hour distribution, a high number of road traffic crashes
were registered from 16:01–17:00. Concomitantly, in the 3-hour distribution in Table 10
below, the maximum crash was registered at 15:01–18:00. As a result, this study considers
16:01–17:00 as a peak hour for road traffic crashes at all road geometric formations.
Table 10. Cross Tabulation of Road Geometric Formation and 3-hour Distribution of Traffic Crash.
Intersection
Horizontal
Curve
Straight
Road
Others
Total
0:00–3:00
3:01–6:00
6:01–9:00
9:01–12:00
12:01–15:00
15:01–18:00
18:01–21:00
21:01–24:00
88
126
874
982
993
1197
750
309
47
40
80
101
121
153
115
69
282
265
1625
1895
1925
2577
1576
680
9
1
12
15
24
37
24
14
426
432
2591
2993
3063
3964
2465
1072
Total
5319
726
10,825
136
17,006
3.13. Road Traffic Crash Frequency and Number of Lane
According to Table 11, a high number of road traffic crashes were observed at one-lane
road geometric formations that accounted for more than 55.85%.
Table 11. Number of lane and Road Traffic Crash Frequency.
Frequency
Percent
9483
4459
2791
273
17,006
55.8
26.2
16.4
1.6
100.0
One Lane
Two Lane
Three and Above Lane
Others
Total
3.14. Road Traffic Crash Outcome and Number of Lane
Table 12 shows that one-lane road geometry formations had the highest number of
road traffic deaths and injuries. It shows that as the number of lanes increased, there was a
reduction in the number of crashes and their outcome. So, road with three lanes and above
has minor role in the occurrence and outcome of the crash.
Table 12. Cross Tabulation of Number of Lane and Road Traffic Crash outcome.
Number of Lane
Crash
Outcome
Fatality
Serious Injuries
Slight Injuries
Total
One
Lane
Two
Lane
Three and
Above Lane
Others
Total
95
2281
7107
65
999
3395
48
624
2119
8
95
170
216
3999
12,791
9483
4459
2791
273
17,006
3.15. Interaction of Road Geometric Formation and Number of Lane on Traffic Crash
Table 13 below shows that although the number of road traffic crashes on a one-lane
road is high, in all lane formations, a high number of road crashes and their outcomes were
registered at the straight and intersection parts of the road geometry formation.
Sustainability 2022, 14, 8475
14 of 25
Table 13. Road Geometric Formation and Number of Lane for the Occurrence of Traffic Crash.
One Lane
Two Lane
Three and Above Lane
Others
Total
Intersection
Horizontal Curve
Straight Road
Others
3374
451
5656
2
1284
181
2992
2
633
84
2072
2
28
10
105
130
5319
726
10,825
136
Total
9483
4459
2791
273
17,006
3.16. Number of Lane and Severity Level of Road Traffic Crash
Based on equation 1 stated above, this study tried to analyze the severity level of the
road lane number based on the number of road traffic fatalities and injuries registered.
Table 14 shown above indicates that even if the number of road traffic crash outcomes
varies according to the number of lanes, a high level of severity in terms of fatalities, serious
injuries, and slight injuries was observed at three and above, one-, and two-lane roads,
respectively. Even if the number of deaths and slight injuries is high on one-lane roads, the
level of severity is high at three-lane and two-lane road geometry configurations.
Table 14. Number of Lane and Severity Level of Road Geometric Formation.
One Lane
Two Lane
Three Lane
and Above
Others
SI @ OL
SI @ TL
SI @ >THL
SI @ O
Slight Injuries
Serious Injuries
Fatality
7107
2281
95
3395
999
65
2119
624
48
170
95
8
0.749
0.241
0.010
0.761
0.224
0.015
0.759
0.224
0.017
0.623
0.348
0.029
Total
9483
4459
2791
273
Where; SI—Severity Index, @-at, OL—One Lane, TL—Two Lane, THL—Three Lane and above, O—others
(Unknowns).
3.17. Causes and Hourly Distribution of Road Traffic Crash
As shown in Table A4 below, a high number of road traffic crashes were registered
due to the improper use of traffic control devices such as traffic signs, signals, and marks.
In the 3-hour distribution, a high road traffic crash happened at 15:01–18:00 due to the
improper use of a road traffic control device. In addition to that, a high concentration of
road traffic crashes was registered at 16:01–17:00 based on a 1-hour distribution, and the
study considers this time as the peak hour for the occurrences of road traffic crashes.
3.18. Causes and Road Traffic Crash Outcome
Road traffic accidents such as fatalities, serious injuries, and slight injuries are the
outcomes of road traffic crashes that happen for different reasons. Based on Table A5
below, the study indicated that the maximum number of deaths and injuries registered
were caused by the improper use of road traffic signs.
3.19. Frequency of Road Traffic Crash and Responsible Body
As shown in Table 15, different participants contributed to the occurrences of the road
traffic crash to varying degrees. This study indicated that drivers contributed more than
82.6% of the road traffic crash frequency in the study area. This shows the driver was
responsible for the death and injuries of road users. Furthermore, pedestrians also play a
significant role in the occurrence of road traffic accidents.
Sustainability 2022, 14, 8475
15 of 25
Table 15. Road traffic Crash Frequency and Responsible Body.
Frequency
Percent
Driver
Passenger
Pedestrian
Failure of Traffic Control Devices
Vehicular Failure
Others
14,041
76
1803
10
1056
20
82.6
0.4
10.6
0.1
6.2
0.1
Total
17,006
100.0
As indicated above, most road traffic crashes happen due to the improper use of road
traffic signs by drivers and pedestrians. As a result, it was advisable for the stakeholders to
train road users, mostly drivers, how to properly use road traffic control devices and apply
law enforcement that is used to minimize the occurrences of road traffic crashes and their
outcomes in the study area.
3.20. Road Traffic Crash Outcome and Responsible Body
Table 16, shown below, indicates that drivers can contribute to a huge loss of life and
injuries. As a result, the driver bears a large portion of the blame for road traffic accidents
that result in significant loss of life and physical harm to other road users.
Table 16. Cross Tabulation of Responsible Body and Road Traffic Crash Outcome.
Fatality
Serious Injuries
Slight Injuries
Total
0
16
7
202
13
838
20
1056
Others
Vehicular Failure
Failure of Traffic
Control Devices
Pedestrian
Passenger
Driver
0
1
9
10
87
0
113
621
29
3139
1095
47
10,789
1803
76
14,041
Total
216
3999
12,791
17,006
3.21. Road Geometric Formation and Responsible Body in Traffic Crash
Table 17, shown below, indicates that the driver was highly responsible for the occurrence of road traffic crashes on all parts of the road network. In addition to driver problems
and errors, vehicular failure plays a vital role in the occurrence of road traffic crashes at
the intersection parts of the road network. Pedestrians also play a significant role in the
occurrence of road traffic crashes. So, drivers and pedestrians have a significant role in the
occurrences of road traffic crashes at the stated road geometric formation.
Table 17. Cross Tabulation of Road Geometric Formation and Responsible Body in Traffic Crash.
Intersection
Horizontal
Curve
Straight
Road
Others
Total
1
740
4
11
15
304
0
1
20
1056
Others
Vehicular Failure
Failure of Traffic
Control Devices
Pedestrian
Passenger
Driver
3
2
4
1
10
169
3
4403
42
1
666
1556
71
8875
36
1
97
1803
76
14,041
Total
5319
726
10,825
136
17,006
Sustainability 2022, 14, 8475
16 of 25
4. Conclusions and Recommendation
Road geometric formation has an impact on the occurrences of road traffic crashes
and their outcomes. The aim of this study was to analyze the impact of road geometric
formation on road traffic crashes and their outcome. For further analyze the impact of
road geometric formation on road traffic crash and its outcome, the study used Budapest
city road traffic crash data based on the nature, convenience, and availability. The study
used tools such as Microsoft Excel, the Statistical Package for Social Science (SPSS-20),
and Quantum Geographic Information System (Q-GIS) for data organization and analysis.
Multinomial logistic (MNL) regression is used to identify determinants of road traffic
crashes. The Severity Index (SI) is used to rate the level of severity of determinant factors.
A Multilayer Perceptron Artificial Neural Network (MLP-ANN) is used to analyze the
determinant factors and the impacts of road geometric formation. The study used both
inferential and descriptive statistics.
Multicollinearity tests, p-value, overdispersion, and other statistical model testing
were undertaken to analyze the significance of the data and variable for modeling and
analysis. This study considers a variable which had a correlation coefficient less than
80%. The maximum variable similarity was observed between light conditions and hourly
distribution with a value of 28.9%. Alcohol consumption and collision type were also
responsible for 24.5% of the similarities. As a result, all variables were used for analysis
and modeling.
Both dependent and independent variables for model selection were undertaken with
an overdispersion test by comparing the mean and variance of the data. The analysis
indicated that more than 80% of the data was not overdispersed. As a result, multinomial
logistic regression was selected for analysis determinant factor. In addition, the p-value
was used to determine the level of accuracy. The variable used for model development was
considered with a p-value of 0.05. This means the accuracy level of the analysis was more
than 95%.
In Multilayer Perceptron Artificial Neural Network analysis, the percentage of incorrect predictions both for testing and training was analyzed. The analysis indicated that
the percentage of incorrect predictions was less than 25%, which shows the accuracy level
of the model was 75% and above. So, the model was a good fit with some training and
testing errors.
The result of the study indicated that a high frequency of road traffic crashes and
their outcomes are observed in straight road geometric formations that account for around
63.7%. The horizontal curve of the road is a severe road geometric. In addition to that, a
high number of road traffic crashes and their outcomes were registered on one-lane roads.
However, the severity level was high at three-lanes and above.
The Multinomial Logistic (MNL) Regression Model output indicated that light conditions, collision type, alcohol consumption, speed limit, and road geometric formation
have significant impacts on the occurrences of road traffic crashes and their outcomes. In
terms of collisions, a high number of road traffic deaths and injuries were registered due to
collisions happening between vehicles and vehicles with pedestrians. Vehicle collisions
(especially rear-end collisions) play an important role in the occurrence of road traffic
crashes at straight and intersection parts of the road geometry.
According to the Multilayer Perceptron Artificial Neural Network (MLP-ANN), road
horizontal curved geometry has a positive and significant relationship with road traffic
fatalities. The intersection and horizontal curve had positive interactions and a strong
impact on the occurrences of slight injuries. Concomitantly, the one-lane road has positive
and strong interaction with road traffic and causes serious and slight injuries. Meanwhile,
two-lane roads had a positive and strong interaction with road traffic and slight injuries. In
this case, one-lane and three-lane roads have a positive impact on the number of fatalities
on the road.
While comparing the statistical analysis results of the models stated above, even if in
MNL-regression model, road geometry was not a determinant factor for fatal road traffic
Sustainability 2022, 14, 8475
17 of 25
accidents, the MLP-ANN model indicates that intersection road geometric formation has
a strong and positive impact on the occurrences of fatal road traffic accidents. While the
severity index analysis indicated that the highest severity value was registered on the
horizontal curvature of road geometry, in contrast, the MLP-ANN showed that the impact
of horizontal curve road geometry has a strong and positive impact on the incidence of
slight injuries. On three-lane and above roads, which are highly severe road networks in
this study, the finding of the MLP-ANN model supports that three-lane above-roads had a
slightly strong and positive impact on the occurrences of fatal road traffic accidents.
A high number of road traffic crashes were registered due to the improper use of
traffic control devices. The primary reasons for the occurrences of a road traffic crash at an
intersection, horizontal curve, and straight road geometric formation were the improper
use of road traffic signs; road pavement condition, and stopping sight distance problems,
respectively. For the occurrences of road traffic crashes and their outcomes, 16:01–17:00
was a peak hour at all geometries of the road. In this study, the driver played a vital role in
the occurrences of road traffic crashes and their outcomes accounted for more than 82.6%.
This shows the driver was responsible for the occurrences of road traffic crashes and their
outcomes in all of the road geometric formations. Since most road traffic crashes and their
outcomes happen due to the improper use of road traffic signs, this study recommends
and advises stakeholders to train road users, primarily drivers, on how to properly use
road traffic control devices and respect rules and regulations to reduce the occurrences and
outcomes of road traffic crashes.
In addition to that, to overcome this problem, further investigation is needed to
understand the reason behind why road users were not properly using traffic control
devices, and detailed remedial action must be undertaken on traffic control device usage.
In general, the stakeholders must emphasize the above-indicated determinant factors to
mitigate the problem.
Author Contributions: The first author (D.J.) of this study contribution was gap identification,
literature review, data collection and organization, result analysis and preparation of draft paper.
The co-author (T.S.) followed up over all activity of the research work, review and comments,
participated in result analysis work, etc. All authors have read and agreed to the published version
of the manuscript.
Funding: This work was supported by the Tempus Public Foundation (Hungary) within the framework of Stipendium Hungaricum Scholarship.
Informed Consent Statement: Not Applicable.
Data Availability Statement: The data presented in this study are available on request from the
corresponding author. The data are not publicly available due to privacy.
Acknowledgments: It gives us great pleasure to honor those who contributed their precious time
in reviewing and commenting on the report while conducting this research article. The research
was supported by OTKA-K20-134760—Heterogeneity in user preferences and its impact on transport project appraisal led by Adam TOROK and by OTKA-K21-138053—Life Cycle Sustainability
Assessment of road transport technologies and interventions by Mária Szalmáné Dr. Csete.
Conflicts of Interest: The authors declare no conflict of interest.
Sustainability 2022, 14, 8475
18 of 25
Appendix A
Table A1. Road Traffic Crash District Distribution.
Budapest District
Frequency
Percent
435
819
903
592
553
578
583
987
861
1089
1312
401
1353
1442
694
456
669
701
516
581
583
434
329
135
17,006
2.6
4.8
5.3
3.5
3.3
3.4
3.4
5.8
5.1
6.4
7.7
2.4
8.0
8.5
4.1
2.7
3.9
4.1
3.0
3.4
3.4
2.6
1.9
0.8
100.0
District I
District II
District III
District IV
District V
District VI
District VII
District VIII
District IX
District X
District XI
District XII
District XIII
District XIV
District XV
District XVI
District XVII
District XVIII
District XIX
District XX
District XXI
District XXII
District XXIII
Unknown
Total
Table A2. Cross Tabulation of Hourly Distribution of Road Traffic Crash and Geometric Formation.
Road Geometry Formation
Intersection Horizontal
Curve
1-hour Distribution.
0:00–1:00
1:01–2:00
2:01–3:00
3:01–4:00
4:01–5:00
5:01–6:00
6:01–7:00
7:01–8:00
8:01–9:00
9:01–10:00
10:01–11:00
11:01–12:00
12:01–13:00
13:01–14:00
14:01–15:00
15:01–16:00
16:01–17:00
17:01–18:00
18:01–19:00
19:01–20:00
20:01–21:00
21:01–22:00
22:01–23:00
23:01–24:00
Total
Straight
Road
Others
Total
37
25
26
30
25
71
190
333
351
320
351
311
333
348
312
362
431
404
339
238
173
133
108
68
21
19
7
16
11
13
23
30
27
33
28
40
40
49
32
45
59
49
53
34
28
24
28
17
126
84
72
64
61
140
358
620
647
629
613
653
619
653
653
751
950
876
667
549
360
301
247
132
1
6
2
0
0
1
3
4
5
6
6
3
5
9
10
9
17
11
13
8
3
4
6
4
185
134
107
110
97
225
574
987
1030
988
998
1007
997
1059
1007
1167
1457
1340
1072
829
564
462
389
221
5319
726
10,825
136
17,006
Sustainability 2022, 14, 8475
19 of 25
Table A3. Cross Tabulation of Primary Reason for Road Traffic Crash and Geometric Formation.
Road Geometry Formation
Primary
Reason
Failure of Brake
Careless Driving
Chassis Failure
Disruption of Oncoming Vehicle
Driver Distraction
Road Crossing Disturbing Behaviors
Early Starting of Vehicle
Engine Failure
Entering to the Market
Failure to Illuminate Vehicle
Glare with Reflector
Animal
Improper use of Road Sign
Improper use of Lane
In attention during Takeoff and Landing
Invisibility at Bend and Bump
Irregular Evasion
Irregular Lane Change
Irregular Reversal
Irregular Transport of Passenger and Goods
Irregular Turn
Jumping in or out of the vehicle
Lack of Side Spacing
Leaving Child an attended
Malaise
Obstructing Straight line Traffic
Overloading
Overtaking another Vehicle
Passing in front of Stationery Object
Passing through Prohibited Place
Non-Priority for Electric Vehicle
Non-priority for Pedestrian
Road Pavement Condition
Rubber Defect
Traffic Signal Failure
Traffic Signal Negligence
Sleeping
Slipperiness
Over speed
Steering Failure
Stopping Sight Distance
Vehicle Technical Problem
Traffic Condition
Vehicle using Distractive Sign
Violation of Left Turn Rule
Violation of Right Turn Rule
Weather and Visibility
Others
Total
Total
Intersection
Horizontal
Curve
Straight
Road
Others
2
84
1
4
0
2
5
0
40
0
0
1
1974
10
0
0
2
49
11
1
18
0
2
1
4
124
61
0
4
20
37
437
103
0
2
734
1
2
7
0
123
3
41
18
757
550
9
75
2
31
0
3
0
0
2
0
7
0
1
2
27
8
0
4
1
23
3
0
3
0
7
0
4
23
26
0
4
10
0
32
350
0
1
7
6
1
4
0
33
2
9
0
22
4
37
27
1
788
2
18
3
49
26
2
384
1
0
9
771
44
13
0
31
644
243
3
143
2
104
11
32
326
388
2
188
256
37
1126
1428
7
6
253
36
3
17
4
1712
17
314
23
522
162
134
540
0
23
0
0
0
2
3
0
11
0
0
0
4
0
0
0
0
1
17
0
0
0
0
0
0
2
11
0
1
8
0
2
25
0
0
0
0
0
0
0
1
2
1
0
0
1
0
21
5
926
3
25
3
53
36
2
442
1
1
12
2776
62
13
4
34
717
274
4
164
2
113
12
40
475
486
2
197
294
74
1597
1906
7
9
994
43
6
28
4
1869
24
365
41
1301
717
180
663
5319
726
10,825
136
17,006
Sustainability 2022, 14, 8475
20 of 25
Table A4. Cross Tabulation of Primary Reason for Road Traffic Crash and 3-hour Distributions.
3-Hour Dist.
Primary
Reason
Total
0:00–3:00
3:01–6:00
6:01–9:00
9:01–12:00
12:01–15:00
15:01–18:00
18:01–21:00
21:01–24:00
Failure of Brake
0
0
2
0
1
0
1
1
5
Careless Driving
25
13
134
147
166
200
169
72
926
Chassis Failure
0
0
0
1
0
2
0
0
3
Disruption of Oncoming Vehicle
1
1
6
1
3
3
7
3
25
Driver Distraction
0
0
0
1
0
1
0
1
3
Road Crossing Disturbing Behaviors
4
3
6
5
7
6
15
7
53
Early Starting of Vehicle
1
0
6
5
7
13
2
2
36
Engine Failure
0
0
0
1
0
0
1
0
2
Entering to the Market
3
7
80
78
92
113
54
15
442
Failure to Illuminate Vehicle
0
0
1
0
0
0
0
0
1
Glare with Reflector
0
0
1
0
0
0
0
0
1
Animal
1
0
3
2
3
1
1
1
12
Improper use of Road Sign
56
73
448
467
529
634
381
188
2776
Improper use of Lane
2
2
9
5
13
16
14
1
62
In attention during Takeoff and Landing
1
0
1
1
0
7
3
0
13
Invisibility at Bend and Bump
0
0
0
0
2
0
1
1
4
Irregular Evasion
0
0
11
7
5
7
3
1
34
Irregular Lane Change
9
12
135
139
154
142
96
30
717
Irregular Reversal
2
2
30
85
65
59
26
5
274
Irregular Transport of Passenger and Goods
0
0
2
0
1
0
0
1
4
Irregular Turn
2
4
30
30
27
41
21
9
164
Jumping in or out of the vehicle
0
0
0
0
0
0
2
0
2
Lack of Side Spacing
2
2
16
16
25
27
17
8
113
Leaving Child an attended
0
0
1
2
2
7
0
0
12
Malaise
0
0
5
18
10
5
0
2
40
Sustainability 2022, 14, 8475
21 of 25
Table A4. Cont.
3-Hour Dist.
0:00–3:00
3:01–6:00
6:01–9:00
Obstructing Straight line Traffic
6
3
72
Overloading
10
8
55
Overtaking another Vehicle
0
0
0
0
1
Passing in front of Stationery Object
6
1
41
27
33
Passing through Prohibited Place
5
8
52
44
43
Non-Priority for Electric Vehicle
1
0
10
17
15
Non-priority for Pedestrian
18
35
299
274
Road Pavement Condition
172
113
208
0
0
1
Rubber Defect
Total
9:01–12:00
Total
12:01–15:00
15:01–18:00
18:01–21:00
21:01–24:00
87
96
124
62
25
475
102
100
121
59
31
486
0
0
1
2
53
33
3
197
59
57
26
294
16
10
5
74
195
436
268
72
1597
233
302
353
297
228
1906
0
3
2
1
0
7
Traffic Signal Failure
0
0
0
2
4
1
1
1
9
Traffic Signal Negligence
26
43
135
166
181
212
158
73
994
Sleeping
6
11
4
3
3
11
5
0
43
Slipperiness
0
0
1
2
1
1
0
1
6
Over speed
2
0
4
4
5
8
1
4
28
Steering Failure
0
0
1
0
1
1
1
0
4
Stopping Sight Distance
14
24
262
404
363
506
227
69
1869
Vehicle Technical Problem
0
1
2
4
10
5
2
0
24
Traffic Condition
6
12
41
82
71
95
44
14
365
Vehicle using Distractive Sign
0
1
3
11
10
8
4
4
41
Violation of Left Turn Rule
6
22
208
257
228
308
204
68
1301
Violation of Right Turn Rule
4
7
142
133
142
167
99
23
717
Weather and Visibility
8
9
35
24
25
27
26
26
180
Others
27
15
88
106
119
166
92
50
663
426
432
2591
2993
3063
3964
2465
1072
17,006
Sustainability 2022, 14, 8475
22 of 25
Table A5. Cross Tabulation of Road Traffic Crash Outcome and Its Primary Reason.
Crash Outcome
Primary
Reason
Failure of Brake
Careless Driving
Chassis Failure
Disruption of Oncoming Vehicle
Driver Distraction
Road Crossing Disturbing Behaviors
Early Starting of Vehicle
Engine Failure
Entering to the Market
Failure to Illuminate Vehicle
Glare with Reflector
Animal
Improper use of Road Sign
Improper use of Lane
In attention during Takeoff and Landing
Invisibility at Bend and Bump
Irregular Evasion
Irregular Lane Change
Irregular Reversal
Irregular Transport of Passenger and Goods
Irregular Turn
Jumping in or out of the vehicle
Lack of Side Spacing
Leaving Child an attended
Malaise
Obstructing Straight line Traffic
Overloading
Overtaking another Vehicle
Passing in front of Stationery Object
Passing through Prohibited Place
Non-Priority for Electric Vehicle
Non-priority for Pedestrian
Road Pavement Condition
Rubber Defect
Traffic Signal Failure
Traffic Signal Negligence
Sleeping
Slipperiness
Over speed
Steering Failure
Stopping Sight Distance
Vehicle Technical Problem
Traffic Condition
Vehicle using Distractive Sign
Violation of Left Turn Rule
Violation of Right Turn Rule
Weather and Visibility
Others
Total
Fatality
Serious
Injuries
Slight
Injuries
Total
0
26
0
2
0
3
0
0
5
0
0
0
38
0
0
0
1
6
4
0
3
0
0
0
3
3
0
0
2
17
0
23
20
0
0
16
0
0
5
0
6
0
2
0
11
2
1
17
3
298
0
8
3
17
6
0
95
0
0
5
589
20
2
2
5
141
79
1
48
1
30
0
6
118
145
0
62
102
19
529
550
2
0
194
8
1
6
0
188
9
85
6
286
116
31
183
2
602
3
15
0
33
30
2
342
1
1
7
2149
42
11
2
28
570
191
3
113
1
83
12
31
354
341
2
133
175
55
1045
1336
5
9
784
35
5
17
4
1675
15
278
35
1004
599
148
463
5
926
3
25
3
53
36
2
442
1
1
12
2776
62
13
4
34
717
274
4
164
2
113
12
40
475
486
2
197
294
74
1597
1906
7
9
994
43
6
28
4
1869
24
365
41
1301
717
180
663
216
3999
12,791
17,006
Sustainability 2022, 14, 8475
23 of 25
Table A6. Pearson Correlation Coefficient (Matrix).
Correlation
3-Hour Distribution
Budapest
District
Crash
Outcome
Weather
Condition
Light
Condition
Collusion
Types
Primary
Reason
Road
Geometry
Alcohol
Consumption
Specific
Reason
Pavement
Surface
Number
of Lane
Speed
Limit
3-hour Dist.
Budapest District
Crash Outcome
Weather Condition
Light Condition
Collusion Types
Primary Reason
Road Geometry
Alcohol Consumption
Specific Reason
Pavement Surface
Number of Lane
Speed limit
1
−0.008
−0.003
−0.031 **
0.289 **
0.027 **
−0.006
0.013
0.031 **
0.023 **
0.014
0.014
−0.017 *
−0.008
1
−0.003
−0.008
−0.015
−0.024 **
0.001
−0.073 **
−0.040 **
−0.064 **
−0.003
−0.131 **
0.142 **
−0.003
−0.003
1
−0.003
−0.050 **
−0.105 **
0.054 **
−0.060 **
−0.071 **
−0.057 **
−0.034 **
−0.011
−0.017 *
−0.031 **
−0.008
−0.003
1
−0.140 **
−0.009
0.018 *
0.004
0.023 **
0.002
0.009
0.003
0.012
0.289 **
−0.015
−0.050 **
−0.140 **
1
0.069 **
−0.001
0.036 **
−0.081 **
0.050 **
0.014
0.042 **
0.012
0.027 **
−0.024 **
−0.105 **
−0.009
0.069 **
1
0.004
0.242 **
0.245 **
0.150 **
−0.005
0.039 **
−0.026 **
−0.006
0.001
0.054 **
0.018 *
−0.001
0.004
1
0.026 **
−0.177 **
−0.088 **
0.002
0.054 **
0.055 **
0.013
−0.073 **
−0.060 **
0.004
0.036 **
0.242 **
0.026 **
1
0.053 **
−0.077 **
0.073 **
0.171 **
0.080 **
0.031 **
−0.040 **
−0.071 **
0.023 **
−0.081 **
0.245 **
−0.177 **
0.053 **
1
0.218 **
−0.006
0.045 **
−0.017 *
0.023 **
−0.064 **
−0.057 **
0.002
0.050 **
0.150 **
−0.088 **
−0.077 **
0.218 **
1
0.001
0.164 **
−0.039 **
0.014
−0.003
−0.034 **
0.009
0.014
−0.005
0.002
0.073 **
−0.006
0.001
1
0.085 **
−0.040 **
0.014
−0.131 **
−0.011
0.003
0.042 **
0.039 **
0.054 **
0.171 **
0.045 **
0.164 **
0.085 **
1
0.243 **
−0.017 *
0.142 **
−0.017 *
0.012
0.012
−0.026 **
0.055 **
0.080 **
−0.017 *
−0.039 **
−0.040 **
0.243 **
1
Where, **. Correlation is significant at the 0.01 level (2-tailed). *. Correlation is significant at the 0.05 level (2-tailed).
Sustainability 2022, 14, 8475
24 of 25
References
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
Rahil, A.E.H.M.; Kamel, M.A.; Hamouda, E.A.; Alsulami, B.T. Effect of geometric elements on Accidents’ Rates. Int. J. Eng. Res.
Dev. 2014, 10, 71–75.
Banasode, S. Elements of Highway Geometric Design. India: Firstalert.in 2020. Available online: https://firstalert.in/2020/06/13
/elements-of-highway-geometric-design (accessed on 12 April 2022).
Shanmuk, H. Geometric Design of Highway and Element of Cross Section. Transp. Eng. 2018. Available online: https:
//knowledge4civil.wordpress.com/2016/12/12/geometric-design-of-highway/ (accessed on 25 April 2022).
Pembuain, A.; Priyanto, S.; Suparma, L.B. The Effect of Road Infrastructure on Traffic Accidents. In Proceedings of the 11th Asia
Pacific Transportation and the Environment Conference (APTE 2018), Malang, Indonesia, 18–19 October 2018; Advances in Engineering
Research; Atlantic Press: Dordrecht, The Netherlands, 2018; Volume 186.
Jurewicz, C.; Aumann, C.; Bradshaw, C.; Beesley, R.; Lim, A. Road Geometry Study for Improved Rural Safety; The National
Academies of Sciences, Engineering and Medicine: Washington, DC, USA, 2015; ISBN 978-1-925294-42-2.
Ahmed, I. Road Infrastructure and Road Safety; Transport and Communications Bulletin for Asia and the Pacific: PhD, Associate
Professor, Faculty of Civil Engineering; Transportation Research Alliance (TRA), University Technology Malaysia (UTM): Johor
Bahru, Malaysia, 2013; pp. 1–7.
Polus, A.; Pollatschek, M.A.; Farah, H. Impact of Infrastructure Characteristics on Road Crashes on Two-Lane Highways. Traffic
Inj. Prev. 2005, 6, 240–247. [CrossRef] [PubMed]
Opoku, G. The Influence of Road Infrastructure on Road Traffic Accidents in Ghana. Texila Int. J. Public Health 2019, 2, 2520–3134.
Kustra, W.; Michalski, L. Tools for road infrastructure safety management in Poland. MATEC Web Conf. 2017, 122, 02008. [CrossRef]
Savolainen, P.; Mannering, F. Horizontal and Vertical Curve Impact on Road Accident; Texas Transportation Institute:
Houston, TX, USA, 2007.
Ewan, L.; Al-Kaisy, A.; Hossain, F. Safety Effects of Road Geometry and Roadside Features on Low-Volume Roads in Oregon; Montana
State University: Bozeman, MT, USA, 2016; Volume 2580, pp. 47–55.
Abebe, M.T. Quantifying the Influence of Road Geometric Parameters on Road Safety (Case Study: Hawassa-Shashemene-Bulbula
Rural Two-Lane Highway, Ethiopia). J. Transp. Technol. 2019, 9, 354–380. [CrossRef]
Othman, S.; Thomson, R.; Lannér, G. Identifying critical road geometry parameters affecting crash rate and crash type. In
Proceedings of the Annals of Advances in Automotive Medicine/Annual Scientific Conference, Association for the Advancement
of Automotive Medicine. Gothenburg, Sweden, 5 October 2009; Volume 53, pp. 155–165.
Othman, S.; Thomson, R.; Lannér, G. Are driving and overtaking on right curves more dangerous than on left curves? Ann. Adv.
Automot. Med. 2010, 54, 1943–2461.
Kamat, S.R.; NurulHuda, A.; Dan, M.P.; Mohamad, M. CarAccidents Analysis Through Virtual Window Interface. In
International Conference on Design and Concurrent Engineering, iDECON; Universiti Teknikal Malaysia Melaka (UTeM):
Durian Tunggal, Malaysia, 2012.
Kriswardhana, W.; Hasanuddin, A.; Palestine, I.M. Modelling Road Traffic Accident Rate and Road Geometric Parameters Relationship;
AIP: Jember, Indonesia, 2020; p. 020003.
Miche, M.; Francois, W.J.; George, N.; Gilles, W. Impact of road geometric design elements on road traffic accidents in the city of
Yaoundé Cameroon. Int. J. Eng. Innov. Technol. (IJEIT) 2020, 10, 2277–3754.
Tay, R.; Rifaat, S.M. Factors contributing to the severity of intersection crashes. J. Adv. Transp. 2010, 41, 0197–6729. [CrossRef]
U.S. Department of Transportation. Crash Factors in Intersection-Related Crashes: An On-Scene Perspective; National Highway Trafic
Safety Administration: Washington, DC, USA, 2010.
Abbasi, M.; Piccioni, C.; Sierpiński, G.; Farzin, I. Analysis of Crash Severity of Texas Two Lane Rural Roads Using Solar Altitude
Angle Based Lighting Condition. Sustainability 2022, 14, 1692. [CrossRef]
Hosseinpour, M.; Yahaya, A.S.; Sadullah, A.F. Exploring the effects of roadway characteristics on the frequency and severity of
head-on crashes: Case studies from Malaysian Federal Roads. Accid. Anal. Prev. 2014, 62, 209–222. [CrossRef] [PubMed]
Anarkooli, A.J.; Persaud, B.; Hosseinpour, M.; Saleem, T. Comparison of univariate and two-stage approaches for estimating crash frequency by severity-Case study for horizontal curves on two-lane rural roads. Accid. Anal. Prev. 2019, 129,
382–389. [CrossRef] [PubMed]
Starkweather, J.; Moske, A.K. Multinomial Logistic Regression; The University of North Texas: Denton, TX, USA, 2011. Available
online: https://it.unt.edu/sites/default/files/mlr_jds_aug2011.pdf (accessed on 25 April 2022).
Abdulhafedh, A. Incorporating the Multinomial Logistic Regression in Vehicle Crash Severity Modeling: A Detailed Overview. J.
Transp. Technol. 2017, 7, 279. [CrossRef]
TRM Mgmt Group. VA Technical Reference Model v 22.4. Available online: https://www.oit.va.gov/Services/TRM/ToolPage.
aspx?tid=7743 (accessed on 29 May 2022).
Mesquitela, J.; Elvas, L.B.; Ferreira, J.C.; Nunes, L. Data Analytics Process over Road Accidents Data—A Case Study of Lisbon
City. ISPRS Int. J. Geo-Information 2022, 11, 143. [CrossRef]
Flenniken, J.M.; Stuglik, S.; Iannone, B.V. Quantum GIS (QGIS): An Introduction to a Free Alternative to More Costly GIS Platforms;
University of Florida: Gainesville, FL, USA, 2020.
Sustainability 2022, 14, 8475
28.
29.
30.
31.
32.
33.
34.
35.
36.
37.
38.
39.
40.
25 of 25
Shaira, H.; Naik, P.R.; Pracheth, R.; Nirgude, A.S.; Nandy, S.; Hiba, M.M.; Karthika, S. Epidemiological profile and mapping
geographical distribution of road traffic accidents reported to a tertiary care hospital, Mangaluru using quantum geographic
information system (QGIS). J. Fam. Med. Prim. Care 2020, 9, 3652.
Razzak, J.; Khan, U.R.; Jalal, S. Application of geographical information system (GIS) for mapping road traffic injuries using
existing source of data in Karachi, Pakistan—A pilot study. J. Pak. Med. Assoc. 2011, 61, 640.
Mahanta, J. Introduction to Neural Networks, Advantages and Applications. Data Sci. 2017, 13. Available online:
https://towardsdatascience.com/introduction-to-neural-networks-advantages-and-applications-96851bd1a207 (accessed on
17 April 2022).
Greene, W.H. Econometric Analysis; Pearson: Boston, MA, USA, 2012; pp. 810–811. Available online: http://www.mysmu.edu/
faculty/zlyang/ECON6002_21-Web/William_H_Greene-Econometric_Analysis-7th.pdf (accessed on 13 March 2022).
Karl-Olov, H. Road Improvement and Traffic Safety Project; Highways General Directorate: Ankara, Turkey, 2001; p. 6.
Bham, G.H.; Javvadi, B.S.; Manepalli, U.R. Multinomial Logistic Regression Model for Single-Vehicle and Multivehicle Collisions
on Urban U.S. Highways in Arkansas. J. Transp. Eng. 2012, 138, 2473–2893. [CrossRef]
Shiran, G.; Imaninasab, R.; Khayamim, R. Crash Severity Analysis of Highways Based on Multinomial Logistic Regression Model,
Decision Tree Techniques, and Artificial Neural Network: A Modeling Comparison. Sustainability 2021, 13, 5670. [CrossRef]
Abirami, S.; Chitra, P. Multilayer Perceptron. Adv. Comput. 2020, 117. Available online: https://www.sciencedirect.com/topics/
computer-science/multilayer-perceptron (accessed on 25 April 2022).
Al-Saif, A.M.; Abdel-Sattar, M.; Aboukarima, A.M.; Eshra, D.H. Application of a multilayer perceptron artificial neural network
for identification of peach cultivars based on physical characteristics. Peer J. 2021, 9, e11529. [CrossRef]
Ogwueleka, F.N.; Misra, S.; Ogwueleka, T.C.; Fernandez-Sanz, L. An Artificial Neural Network Model for Road Accident Prediction: A Case Study of Khuna Metropolitan City; Khulna University of Engineering & Technology: Khulna, Bangladesh, 2018;
ISBN 978-984-34-3502-6.
de Medrano, R.; Aznarte, J.L. A New Spatio-Temporal Neural Network Approach for Traffic Accident Forecasting. Appl. Artif.
Intell. 2021, 35, 782–801. [CrossRef]
Ghorbani, M.A.; Zadeh, H.A.; Isazadeh, M.; Terzi, O. A comparative study of artificial neural network (MLP, RBF) and support
vector machine models for river flow prediction. Environ. Earth Sci. 2016, 75, 476. [CrossRef]
Abdel-Aty, M.A.; Abdelwahab, H.T. Predicting Injury Severity Levels in Traffic Crashes: A Modeling Comparison. J. Transp. Eng.
2004, 130, 2473–2893. [CrossRef]