The Impact of Road Geometric Formation on Traffic Crash and Its Severity Level

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.

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 cross-section. 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 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 [10,11,12]. 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 LED-raised 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 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 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].

$$Seveity Index \left(SI\right)=\frac{Number of Injuries}{Total Number of Crash} or \frac{Number of Death}{Total Number of Crash}$$

(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) A_i for i = 1, 2 … n. Along with each observation, A_i, there is a set of observed values X₁, …, X_n of explanatory variables. The output A_i 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].

$$A_i\backslash X_i\dots \dots \dots \dots \dots \dots ..X_n, for i=1,2,3\dots \dots .n$$

(2)

Multinomial Logistic Regression is often written in terms of a latent variable model as stated below.

$$\begin{gathered} A_i^{1*}={\beta }_{1*}{X}_i+{\epsilon }_1 \\ {A}_i^{2*}={\beta }_{2*}{X}_i+{\epsilon }_2 \\ \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \dots \\ {A}_i^{n*}={\beta }_{n*}{X}_i+{\epsilon }_n \\ where \\ \epsilon ~N\left(0,{\displaystyle \sum }\right) \end{gathered}$$

Based on the above relationships, the model that helps to predict road traffic accident can be defined as a predicting variable A.

$$A=\beta +{\beta }_1{X}_1+{\beta }_2{X}_2+\dots \dots \dots \dots \dots \dots \dots \dots .+{\beta }_n{X}_n+\epsilon$$

(3)

where: A = Dependent (predicted) variable; β = Constant; β_i = 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 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 [37,38,39].

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. 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.	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 *
Budapest District	−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 **
Crash Outcome	−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 *
Weather Condition	−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
Light Condition	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
Collusion Types	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 **
Primary Reason	−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 **
Road Geometry	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 **
Alcohol Consumption	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 *
Specific Reason	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 **
Pavement Surface	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 **
Number of Lane	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 **
Speed limit	−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).

.

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. Mean and Variance of Variables.

Variable		Number		Mean	Variances
		Valid	Missed
Dependent	Crash Outcome	17,006	0	2.74	0.218
Independent	Light Condition	17,006	0	1.27	0.233
	Collision Type	17,006	0	3.94	9.204
	Road Geometry	17,006	0	2.35	0.900
	Reason	17,006	0	4.62	9.821
	Pavement Surface	17,006	0	1.06	0.119
	Weather Condition	17,006	0	2.66	0.821
	Alcohol Consumption	17,006	0	5.05	0.512
	Speed Limit	17,006	0	2.11	0.218

, 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 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).

3. Result and Discussion

These sections of the study attempt to identify the output of the analysis and discuss the findings. It contains the road traffic crash distribution and its frequency in terms of independent variables, and cross-tabulation between dependent and independent variables to signify the impact of other variables on road traffic crashes and their outcome. Moreover, it cross-tabulates the independent variables against each other. In general, this study tried to bring reliable findings and their implications that define the impact of road geometric formation on traffic crashes and their severity level. For more information, please see the study findings discussed below:

3.1. Road Network and Traffic Crash Distribution

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 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.

3.2. Road Traffic Crash District Distribution and District Level of Concentration

The study site has 23 districts. The study tried to rate the level of a road traffic crash in percentage based on its concentration. As a result, the study grouped the area into three; 0–3%, 3–6%, and 6–9% as high, intermediate, and low, respectively. The maximum and minimum road traffic crash concentrations were observed in District XIV with 8.5% and XXIII with 1.9%, respectively. For more information, see Figure 2 below and

Table A1. Road Traffic Crash District Distribution.

Budapest District
	Frequency	Percent
District I	435	2.6
District II	819	4.8
District III	903	5.3
District IV	592	3.5
District V	553	3.3
District VI	578	3.4
District VII	583	3.4
District VIII	987	5.8
District IX	861	5.1
District X	1089	6.4
District XI	1312	7.7
District XII	401	2.4
District XIII	1353	8.0
District XIV	1442	8.5
District XV	694	4.1
District XVI	456	2.7
District XVII	669	3.9
District XVIII	701	4.1
District XIX	516	3.0
District XX	581	3.4
District XXI	583	3.4
District XXII	434	2.6
District XXIII	329	1.9
Unknown	135	0.8
Total	17,006	100.0

.

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 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. Road Geometry Formation and Traffic Crash Frequency.

	Frequency	Percent
Intersection	5319	31.3
Horizontal Curve	726	4.3
Straight Road	10,825	63.7
Others	136	0.8
Total	17,006	100.0

shown below indicates how road traffic crashes and their outcomes vary with road geometric formation.

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. Cross-tabulation of Crash Outcome and Road Geometric Formation.

		Intersection	Horizontal Curve	Straight Road	Others	Total
Crash Outcome	Slight Injuries	4192	529	7989	81	12,791
	Serious Injuries	1081	177	2693	48	3999
	Fatality	46	20	143	7	216

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.

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. Road Geometric Formation and Its Severity Level.

	Intersection	Horizontal Curve	Straight Road	Others	SI @ I	SI @ HC	SI @ SR
Slight Injuries	4192	529	7989	81	0.788	0.729	0.738
Serious Injuries	1081	177	2693	48	0.203	0.244	0.249
Fatality	46	20	143	7	0.009	0.028	0.013

Where; SI—Severity Index, I—Intersection, HC—Horizontal Curve, SR—Straight Road.

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, 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.

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. 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	−14.340	0.941	232.268	1	0.000
	WC	0.169	0.076	4.917	1	0.027	1.184	1.020	1.375
	LC	0.505	0.125	16.464	1	0.000	1.657	1.298	2.115
	CT	0.118	0.028	17.357	1	0.000	1.125	1.064	1.189
	RG	−0.038	0.085	0.202	1	0.653	0.963	0.816	1.136
	AC	1.216	0.176	47.942	1	0.000	3.374	2.391	4.760
	R	0.064	0.026	6.006	1	0.014	1.066	1.013	1.122
	PS	0.262	0.139	3.559	1	0.059	1.299	0.990	1.705
	VL	0.727	0.090	65.023	1	0.000	2.069	1.734	2.469
Serious Injuries	Intercept	−2.367	0.187	159.628	1	0.000
	WC	0.008	0.020	0.147	1	0.701	1.008	0.969	1.048
	LC	0.126	0.038	11.251	1	0.001	1.135	1.054	1.222
	CT	0.039	0.007	35.082	1	0.000	1.039	1.026	1.053
	RG	0.104	0.020	25.784	1	0.000	1.109	1.066	1.155
	AC	0.031	0.027	1.251	1	0.263	1.031	0.977	1.088
	R	0.058	0.006	85.482	1	0.000	1.060	1.047	1.073
	PS	0.152	0.049	9.831	1	0.002	1.164	1.059	1.281
	VL	0.013	0.040	0.098	1	0.754	1.013	0.936	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).

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.

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.

$$A=\beta +{\beta }_{1}WC+{\beta }_{2}LC+{\beta }_{3}CT+{\beta }_{4}RG+{\beta }_{5}AC+{\beta }_{6}R+{\beta }_{7}PS+{\beta }_{8}VL+\epsilon \left(error\right)$$

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. Cross Tabulation between Collusion Type and Road Traffic Crash Outcome.

		Crash Outcome
		Fatality	Serious Injuries	Slight Injuries	Total
Collusion Types	Collision with Animals	0	5	8	13
	Collision with Pedestrian	122	1394	2574	4090
	Pileup Collision	6	177	1456	1639
	Collision with Object	7	114	291	412
	Head on Collision	27	427	1468	1922
	Side Swipe Collision	1	12	46	59
	Side Impact Collision	4	302	1164	1470
	Rear End Collision	49	1568	5784	7401
Total		216	3999	12,791	17,006

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.

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. Model Summary of Error Computations Both the Training and Testing Samples.

		Road Geometry	Number of Lane
Training	Cross Entropy Error	7355.623	7287.471
	Percent Incorrect Predictions	25.0%	25.0%
Testing	Cross Entropy Error	2993.791	3103.688
	Percent Incorrect Predictions	24.3%	24.2%
Dependent Variable: Crash Outcome

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].

The blue and gray lines in Figure 3 show the positive and negative bond between the dependent and independent variables, whose synaptic weight is >0 and <0, respectively. In this diagram, road geometry, and the number of lanes were input (independent variables), and crash outcome (road traffic accident) was output (dependent variable).

The intersection part of the road has a strong and positive contribution to the occurrences of road traffic crashes that cause fatalities, whereas the horizontal curve of the road 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 accidents.

In addition to that, one-lane and two-lane roads had a strong and positive contribution to the occurrence of road traffic injuries. Simultaneously, three-lane and above-average roads have a significant positive and strong impact on the occurrences of minor injuries. It also has a relatively positive and slightly stronger impact on the occurrences of death. So, this indicates that road geometric formation has an impact on the occurrence of road traffic injuries.

3.9. Road Traffic Crash Level of Severity and Road Geometric Formation

The outcome of the crash analysis indicated that a highly severe accident was registered on a straight road. Based on

Table 8. Cross Tabulation of Road Traffic Crash Outcome and Road Geometric Formation.

	Intersection	Horizontal Curve	Straight Road	Others	Total
Fatality	46	20	143	7	216
Serious Injuries	1081	177	2693	48	3999
Slight Injuries	4192	529	7989	81	12,791
Total	5319	726	10,825	136	17,006

shown below, from total road traffic crash 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.

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. Cross Tabulation of Primary Reason for Road Traffic Crash and Geometric Formation.

		Road Geometry Formation				Total
		Intersection	Horizontal Curve	Straight Road	Others
Primary Reason	Failure of Brake	2	2	1	0	5
	Careless Driving	84	31	788	23	926
	Chassis Failure	1	0	2	0	3
	Disruption of Oncoming Vehicle	4	3	18	0	25
	Driver Distraction	0	0	3	0	3
	Road Crossing Disturbing Behaviors	2	0	49	2	53
	Early Starting of Vehicle	5	2	26	3	36
	Engine Failure	0	0	2	0	2
	Entering to the Market	40	7	384	11	442
	Failure to Illuminate Vehicle	0	0	1	0	1
	Glare with Reflector	0	1	0	0	1
	Animal	1	2	9	0	12
	Improper use of Road Sign	1974	27	771	4	2776
	Improper use of Lane	10	8	44	0	62
	In attention during Takeoff and Landing	0	0	13	0	13
	Invisibility at Bend and Bump	0	4	0	0	4
	Irregular Evasion	2	1	31	0	34
	Irregular Lane Change	49	23	644	1	717
	Irregular Reversal	11	3	243	17	274
	Irregular Transport of Passenger and Goods	1	0	3	0	4
	Irregular Turn	18	3	143	0	164
	Jumping in or out of the vehicle	0	0	2	0	2
	Lack of Side Spacing	2	7	104	0	113
	Leaving Child an attended	1	0	11	0	12
	Malaise	4	4	32	0	40
	Obstructing Straight line Traffic	124	23	326	2	475
	Overloading	61	26	388	11	486
	Overtaking another Vehicle	0	0	2	0	2
	Passing in front of Stationery Object	4	4	188	1	197
	Passing through Prohibited Place	20	10	256	8	294
	Non-Priority for Electric Vehicle	37	0	37	0	74
	Non-priority for Pedestrian	437	32	1126	2	1597
	Road Pavement Condition	103	350	1428	25	1906
	Rubber Defect	0	0	7	0	7
	Traffic Signal Failure	2	1	6	0	9
	Traffic Signal Negligence	734	7	253	0	994
	Sleeping	1	6	36	0	43
	Slipperiness	2	1	3	0	6
	Over speed	7	4	17	0	28
	Steering Failure	0	0	4	0	4
	Stopping Sight Distance	123	33	1712	1	1869
	Vehicle Technical Problem	3	2	17	2	24
	Traffic Condition	41	9	314	1	365
	Vehicle using Distractive Sign	18	0	23	0	41
	Violation of Left Turn Rule	757	22	522	0	1301
	Violation of Right Turn Rule	550	4	162	1	717
	Weather and Visibility	9	37	134	0	180
	Others	75	27	540	21	663
Total		5319	726	10,825	136	17,006

, 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. Cross Tabulation of Collision Type and Road Geometric Formation.

	Intersection	Horizontal Curve	Straight Road	Others	Total
Collision with Animals	1	2	10	0	13
Collision with Pedestrian	668	81	3258	83	4090
Pileup Collision	112	41	1485	1	1639
Collision with Object	21	52	332	7	412
Head on Collision	925	141	850	6	1922
Side Swipe Collision	6	4	49	0	59
Side Impact Collision	1011	18	438	3	1470
Rear End Collision	2575	387	4403	36	7401
Total	5319	726	10,825	136	17,006

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.

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.

3.12. Road Geometric Formation and Traffic Crash Hourly Distribution

Table A2. Cross Tabulation of Hourly Distribution of Road Traffic Crash and Geometric Formation.

		Road Geometry Formation				Total
		Intersection	Horizontal Curve	Straight Road	Others
1-Hour Distribution.	0:00–1:00	37	21	126	1	185
	1:01–2:00	25	19	84	6	134
	2:01–3:00	26	7	72	2	107
	3:01–4:00	30	16	64	0	110
	4:01–5:00	25	11	61	0	97
	5:01–6:00	71	13	140	1	225
	6:01–7:00	190	23	358	3	574
	7:01–8:00	333	30	620	4	987
	8:01–9:00	351	27	647	5	1030
	9:01–10:00	320	33	629	6	988
	10:01–11:00	351	28	613	6	998
	11:01–12:00	311	40	653	3	1007
	12:01–13:00	333	40	619	5	997
	13:01–14:00	348	49	653	9	1059
	14:01–15:00	312	32	653	10	1007
	15:01–16:00	362	45	751	9	1167
	16:01–17:00	431	59	950	17	1457
	17:01–18:00	404	49	876	11	1340
	18:01–19:00	339	53	667	13	1072
	19:01–20:00	238	34	549	8	829
	20:01–21:00	173	28	360	3	564
	21:01–22:00	133	24	301	4	462
	22:01–23:00	108	28	247	6	389
	23:01–24:00	68	17	132	4	221
Total		5319	726	10,825	136	17,006

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. Cross Tabulation of Road Geometric Formation and 3-hour Distribution of Traffic Crash.

	Intersection	Horizontal Curve	Straight Road	Others	Total
0:00–3:00	88	47	282	9	426
3:01–6:00	126	40	265	1	432
6:01–9:00	874	80	1625	12	2591
9:01–12:00	982	101	1895	15	2993
12:01–15:00	993	121	1925	24	3063
15:01–18:00	1197	153	2577	37	3964
18:01–21:00	750	115	1576	24	2465
21:01–24:00	309	69	680	14	1072
Total	5319	726	10,825	136	17,006

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.

3.13. Road Traffic Crash Frequency and Number of Lane

According to

Table 11. Number of lane and Road Traffic Crash Frequency.

	Frequency	Percent
One Lane	9483	55.8
Two Lane	4459	26.2
Three and Above Lane	2791	16.4
Others	273	1.6
Total	17,006	100.0

, a high number of road traffic crashes were observed at one-lane road geometric formations that accounted for more than 55.85%.

3.14. Road Traffic Crash Outcome and Number of Lane

Table 12. Cross Tabulation of Number of Lane and Road Traffic Crash outcome.

		Number of Lane
		One Lane	Two Lane	Three and Above Lane	Others	Total
Crash Outcome	Fatality	95	65	48	8	216
	Serious Injuries	2281	999	624	95	3999
	Slight Injuries	7107	3395	2119	170	12,791
Total		9483	4459	2791	273	17,006

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.

3.15. Interaction of Road Geometric Formation and Number of Lane on Traffic Crash

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	3374	1284	633	28	5319
Horizontal Curve	451	181	84	10	726
Straight Road	5656	2992	2072	105	10,825
Others	2	2	2	130	136
Total	9483	4459	2791	273	17,006

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.

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. 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	7107	3395	2119	170	0.749	0.761	0.759	0.623
Serious Injuries	2281	999	624	95	0.241	0.224	0.224	0.348
Fatality	95	65	48	8	0.010	0.015	0.017	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).

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.

3.17. Causes and Hourly Distribution of Road Traffic Crash

As shown in

Table A4. Cross Tabulation of Primary Reason for Road Traffic Crash and 3-hour Distributions.

		3-Hour Dist.								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
Primary Reason	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
	Obstructing Straight line Traffic	6	3	72	87	96	124	62	25	475
	Overloading	10	8	55	102	100	121	59	31	486
	Overtaking another Vehicle	0	0	0	0	1	0	0	1	2
	Passing in front of Stationery Object	6	1	41	27	33	53	33	3	197
	Passing through Prohibited Place	5	8	52	44	43	59	57	26	294
	Non-Priority for Electric Vehicle	1	0	10	17	15	16	10	5	74
	Non-priority for Pedestrian	18	35	299	274	195	436	268	72	1597
	Road Pavement Condition	172	113	208	233	302	353	297	228	1906
	Rubber Defect	0	0	1	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
Total		426	432	2591	2993	3063	3964	2465	1072	17,006

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. Cross Tabulation of Road Traffic Crash Outcome and Its Primary Reason.

		Crash Outcome			Total
		Fatality	Serious Injuries	Slight Injuries
Primary Reason	Failure of Brake	0	3	2	5
	Careless Driving	26	298	602	926
	Chassis Failure	0	0	3	3
	Disruption of Oncoming Vehicle	2	8	15	25
	Driver Distraction	0	3	0	3
	Road Crossing Disturbing Behaviors	3	17	33	53
	Early Starting of Vehicle	0	6	30	36
	Engine Failure	0	0	2	2
	Entering to the Market	5	95	342	442
	Failure to Illuminate Vehicle	0	0	1	1
	Glare with Reflector	0	0	1	1
	Animal	0	5	7	12
	Improper use of Road Sign	38	589	2149	2776
	Improper use of Lane	0	20	42	62
	In attention during Takeoff and Landing	0	2	11	13
	Invisibility at Bend and Bump	0	2	2	4
	Irregular Evasion	1	5	28	34
	Irregular Lane Change	6	141	570	717
	Irregular Reversal	4	79	191	274
	Irregular Transport of Passenger and Goods	0	1	3	4
	Irregular Turn	3	48	113	164
	Jumping in or out of the vehicle	0	1	1	2
	Lack of Side Spacing	0	30	83	113
	Leaving Child an attended	0	0	12	12
	Malaise	3	6	31	40
	Obstructing Straight line Traffic	3	118	354	475
	Overloading	0	145	341	486
	Overtaking another Vehicle	0	0	2	2
	Passing in front of Stationery Object	2	62	133	197
	Passing through Prohibited Place	17	102	175	294
	Non-Priority for Electric Vehicle	0	19	55	74
	Non-priority for Pedestrian	23	529	1045	1597
	Road Pavement Condition	20	550	1336	1906
	Rubber Defect	0	2	5	7
	Traffic Signal Failure	0	0	9	9
	Traffic Signal Negligence	16	194	784	994
	Sleeping	0	8	35	43
	Slipperiness	0	1	5	6
	Over speed	5	6	17	28
	Steering Failure	0	0	4	4
	Stopping Sight Distance	6	188	1675	1869
	Vehicle Technical Problem	0	9	15	24
	Traffic Condition	2	85	278	365
	Vehicle using Distractive Sign	0	6	35	41
	Violation of Left Turn Rule	11	286	1004	1301
	Violation of Right Turn Rule	2	116	599	717
	Weather and Visibility	1	31	148	180
	Others	17	183	463	663
Total		216	3999	12,791	17,006

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. Road traffic Crash Frequency and Responsible Body.

	Frequency	Percent
Driver	14,041	82.6
Passenger	76	0.4
Pedestrian	1803	10.6
Failure of Traffic Control Devices	10	0.1
Vehicular Failure	1056	6.2
Others	20	0.1
Total	17,006	100.0

, 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.

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. Cross Tabulation of Responsible Body and Road Traffic Crash Outcome.

	Fatality	Serious Injuries	Slight Injuries	Total
Others	0	7	13	20
Vehicular Failure	16	202	838	1056
Failure of Traffic Control Devices	0	1	9	10
Pedestrian	87	621	1095	1803
Passenger	0	29	47	76
Driver	113	3139	10,789	14,041
Total	216	3999	12,791	17,006

, 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.

3.21. Road Geometric Formation and Responsible Body in Traffic Crash

Table 17. Cross Tabulation of Road Geometric Formation and Responsible Body in Traffic Crash.

	Intersection	Horizontal Curve	Straight Road	Others	Total
Others	1	4	15	0	20
Vehicular Failure	740	11	304	1	1056
Failure of Traffic Control Devices	3	2	4	1	10
Pedestrian	169	42	1556	36	1803
Passenger	3	1	71	1	76
Driver	4403	666	8875	97	14,041
Total	5319	726	10,825	136	17,006

, 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.

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 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.