Proportion-Based Analytical Hierarchy Process for Determining Prominent Reasons Causing Severe Crashes

Abstract

Governments and authorities worldwide consider road traffic crashes (RTCs) to be a major concern. These crashes incur losses in terms of productivity, property, and life. For a country to establish its road and action plans, it is crucial to comprehend the reasons for and consequences of traffic collisions. The main objective of this research study was to evaluate and rank the important and supporting factors influencing traffic crashes on the road. To identify the most significant accident causation elements, the proportion-based analytic hierarchy process (PBAHP) was used to order the factors in terms of their relative importance. In this study, the city of Al-Ahsa, located in the eastern province of Saudi Arabia, was used as a case study, since this city is the highest RTC-prone area in the region. PBAHP was used to calculate relative importance/weights for different crash types and reasons in terms of their impact on crash severity. It was found that vehicle-overturned collisions which result in fatal crashes have the most weight, whereas “hit motorcycle” crashes result in serious injury crashes. When vehicles (two or more) collide with one another while they are moving, it appears that the likelihood of a fatality in a collision increases. The highest weights for serious injury crashes came from “driver distraction”, “leaving insufficient safe distance”, and “speeding”, which also generated similar and relatively high weights for fatal crashes. Weights from the PBAHP approach were also used to develop utility functions for predicting the severity of crashes. This approach could assist decision-makers in concentrating on the key elements affecting road traffic crashes and enhancing road safety.

1. Introduction

Road traffic crashes (RTCs) cause enormous losses of life and property and are a major concern for public safety worldwide. Road crashes have significant and irreversible economic and social impacts on countries, making them a primary concern for transportation managers globally [1]. According to the World Health Organization (WHO), 1.35 million people die because of RTCs each year, and millions more suffer nonfatal injuries, impairments, and other long-term health effects. Very severe crashes can cause major bodily and monetary harm. Between 1975 and 1998, the number of individuals who were either killed or injured in RTCs increased significantly in Malaysia, Colombia, and Botswana. A study on RTCs in Iran has revealed that the number of people injured in these incidents is about ten times higher than the number of fatalities, with approximately 240,000 instances every year. Given such elevated rates, in addition to the substantial volume of passengers commuting, it is essential to conduct a comprehensive investigation into passenger safety. In Saudi Arabia, RTCs are a major public health problem since they result in large human casualties and material losses. With an average of 7000 RTC fatalities per year, the nation has one of the highest rates in the world [2]. A study that analyzed 404 crash reports from various types of RTCs found that the conditions of the road surroundings were a contributing factor in around 14.5% of all crashes. In 2007, Saudi Arabia saw 30 deaths per 100,000 individuals due to RTCs, resulting in 6358 deaths. According to a report by the WHO, RTCs are the main cause of fatality, injury, and disability in the Kingdom of Saudi Arabia, with a projected cost of SAR 652.5 million to treat those who are injured or killed. According to officials in Saudi Arabia, the country experiences one road traffic crash (RTC) every minute, making it one of the most perilous nations in the world for RTCs. With an average of 21 deaths/day, it is ranked as the second most lethal nation in the Gulf Region and the second deadliest country in the Middle East. These crashes have multiple contributing factors. A thorough investigation of several elements, such as road design, driver behavior, vehicle safety features, traffic volume, and weather conditions, is needed to identify the main causes and factors that result in serious crashes. It is crucial to understand the causes of these crashes in order to implement successful methods to reduce RTCs and enhance road safety in Saudi Arabia.

To determine the reasons behind RTCs in Saudi Arabia, several investigations have been carried out. According to a study by Ansari et al. (2014) [3], driving-related variables are the main reasons for traffic crashes in the nation. The study indicated that the most important driver-related variables causing RTCs in Saudi Arabia were speeding, distracted driving, and reckless driving. The survey also found that key causes of RTCs in the nation included aspects of the road, such as inadequate traffic signs and signals and bad road design. Research into the causes of young Saudi drivers’ participation in RTCs was undertaken by Hassan et al. (2016) [4]. According to the survey, speeding and being late were to blame for 60% of unsafe driving behaviors and more than 70% of fines in Riyadh, Saudi Arabia. A study by Ramisetty-Mikler and Almakadma (2016) [5] on adolescent drivers in Riyadh, Saudi Arabia, revealed that 40% of drivers engage in “drifting” as a recreational activity with their vehicles, despite being aware of the risks involved. In contrast to less experienced and less educated drivers, Issa’s (2016) [6] study indicated that educated and experienced drivers demonstrated a higher likelihood of being engaged in RTCs. Vehicle collisions, multiple-vehicle crashes, and pedestrian collisions are the most common types of crashes in the Al-Ahsa region of Saudi Arabia, according to research by Islam et al. (2022) [7]. Although they occur less frequently, RTCs had a high severity index in the study area, with 73% of severe RTCs involving people between the ages of 15 and 44. The study also discovered that people between the ages of 15 and 44 were more frequently involved in RTCs with a large number of fatalities and injured victims. According to Rahman et al.’s (2022) [8] research, there is a 26% chance of a collision occurring merely as a result of speeding, a rise of 63% from an earlier estimate. Furthermore, the likelihood of a collision rises from 26% to 33% when speeding and brake failure are factored in, more than tripling the initial odds of a collision. Islam et al. (2022) [9] used the random forest approach to evaluate the same research region and discovered that the two most significant factors impacting the severity of injuries in RTCs are the reason for the crash and the kind of collision. They also demonstrated that the research region had substantial spatial dependence, with clustered spatial patterns found within a distance threshold of 500 m, in addition to the findings on the factors determining the severity of injuries in RTCs. The analysis used by the authors, which employed Getis Ord Gi*, the crash severity index, and spatial autocorrelation analysis based on Moran’s I, was successful in locating and ranking crash hotspots as well as evaluating the severity of the crashes.

The Saudi Arabian government has implemented various measures to enhance road safety across the country in recent years, including in the region of Al-Ahsa. These measures include increasing the presence of traffic police personnel, improving the road infrastructure, and intensifying awareness programs aimed at promoting safe driving practices. Additionally, the government has established the Saudi Traffic Safety Society, which plays a significant role in fostering road safety through educational campaigns and awareness initiatives. As a result of these concerted efforts, the number of road crashes in Al-Ahsa and Saudi Arabia as a whole has witnessed a decline in recent years [7]. However, it remains crucial for drivers to exercise caution, comply with traffic regulations, and prioritize the safety of themselves and other road users in order to prevent collisions and ensure a safe driving environment.

The study aims to improve our understanding of severe crashes and offer a more useful analytical framework for analyzing crash data by combining these efforts in the following two main points:

• Providing insight into the types and causes of severe crashes in the Al-Ahsa region:

This study makes a significant addition by offering insightful information about the factors that contribute to serious crashes in the Saudi Arabian province of Al-Ahsa. The study intends to find patterns and trends related to severe crashes using data analysis, taking into account underlying causes such as the type of vehicle, crash types, road conditions, and driver behavior.

2.

Employing the newly proposed proportion-based analytic hierarchy process (PBAHP) for crash data analysis:

The development and use of the proportion-based analytic hierarchy process (PBAHP) approach for crash data analysis is another significant contribution of this study. The study suggests a novel strategy that overcomes the methodological limitation associated with the conventional analytic hierarchy process (AHP) method, such as inconsistent pairwise comparisons and reliance on subjective judgments, by allowing the weighting and ranking of items based on their relative occurrence.

The newly developed PBAHP was applied to analyze the factors involved in traffic crashes and then to develop a utility function for predicting the severity of crashes. The implementation of this technique has not been found in the literature; instead, several variants of AHP were used for the data collected through surveys and interviews. In the proposed approach of PBAHP, the proportions of crashes of different types and those because of different reasons were used to calculate the ranking of these factors. Later, these rankings were summarized to calculate the weight/impact of each factor on the severity of crash, resulting in the formation of the utility function.

This study aimed to use these contributions to develop evidence-based road safety actions with the goal of lowering the frequency and severity of serious crashes in Al-Ahsa, which opens the door for improved road safety initiatives and policies in the study area, with the possibility of wider applications in comparable circumstances.

The subsequent sections of this paper are structured as follows. In Section 2, a comprehensive review of recent literature relevant to this research field is presented, where several variants of AHP methods are compared and the proposed PBAHP method is discussed. Section 3 provides detailed information about the data sources and their characteristics, as well as the steps taken for data preparation. The modeling methodology is outlined in Section 4, where the formulation of the proposed PBAHP is presented. Section 5 presents the results and discussion based on the analysis of the crash data. Finally, in Section 6, the key findings of this study are summarized, along with recommendations for authorities.

2. Literature Review

Many investigations have been made into the types and causes of auto crashes. Crash severity analysis, which focuses on identifying the elements that contribute to serious crashes, is one of the most discussed aspects in this area. The ability of conventional crash severity analysis methodologies to offer a thorough grasp of the intricate interactions between the different components that contribute to severe crashes is constrained. Experts employ a variety of techniques to pinpoint the main causes of catastrophic crashes, as well as the sorts of crashes that cause the most severe effects. These approaches incorporate statistical analysis, professional judgment, and machine learning strategies. RTC analysis and prevention using cutting-edge methods have become urgently necessary. Multicriteria decision-making (MCDM) tools have been used in several research works to examine the causes of traffic crashes [10,11,12,13,14,15,16].

The analytical hierarchy process (AHP) is a multicriteria decision-making tool that enables decision-makers to specify pertinent safety criteria in a methodical manner. Based on the relative relevance of factors such as crash severity, frequency, road conditions, driving habits, and vehicle attributes in connection to the goals of traffic safety, AHP makes it easier to compare and rank them. For instance, Cheng et al. (2011) [17] used the AHP model in a modeling study to investigate hidden safety issues and determine the root causes of traffic accidents in China. The study’s goal was to offer insights into the state of road traffic safety and to suggest preventive measures based on the analysis. Despite the possibility of certain subjective elements being present in the AHP model employed in the study and the possibility that the expert grading may not have been entirely thorough, they ultimately obtained objective materiality levels of road crash causes in China. For the case of Manila, Philippines, Fernandeza et al. (2020) [18] examined how road users’ prioritization of the factors that cause traffic crashes is affected by their understanding of traffic signs, and found that a thorough understanding of traffic signs led to a more precise and trustworthy ranking of the causes of traffic crashes. They compared AHP with the forced ranking method in terms of a pairwise comparison of crash factors and found that the AHP performed better in representing the importance of factors with in-depth details. Saifullizan et al. (2022) [19] used the AHP method to rank and weigh car crashes according to the different categories of injury using data from Balai Traffic Batu Pahat Johor, Malaysia, from October 2016 to October 2020. They prioritized wreckage injuries at the top of the hierarchy system, followed by fatal injuries, minor injuries, and serious injuries. They reported that the suitability of the model and technology were key factors in the successful application of AHP, and recommended enhancing the application of AHP, including via parameter determination, reaching consensus, and framework modification, in order to achieve progress.

Based on the conventional AHP method, an Improved Fuzzy-AHP method was suggested by Hu et al. (2009) [20] for evaluating the safety of road traffic. The importance of each assessment index was determined using the scale technique after providing a summary of the important road traffic safety indices. Utilizing a fuzzy consistent matrix, queuing each index according to relevance, and adopting an accident prevention mechanism in accordance with the index, the weight of each index could be confirmed. In order to compare and quantify the driver behavioral attitudes in different traffic cultures, such as in Hungary, Turkey, Pakistan, and China, Farooq et al. (2020) [21] devised the fuzzy-AHP (FAHP) framework built on a three-level hierarchical structure. Based on pairwise comparisons (PCs) of drivers’ responses to the driver behavior questionnaire (DBQ), the FAHP process calculated the weight factors and rated the significant driver behavior criteria. According to their study’s findings, “violations” were the most important driver behavior standard for level 1 in all nominated regions, with the exception of Hungary. All regions, with the exception of Turkey, observed “aggressive violations” to be the most important driver behavior criterion for level 2. Additionally, level 3 drivers in Hungary and Turkey rated “driving while intoxicated” as the most important driver behavior criterion. Farooq and Moslem (2022) [22] applied the Pythagorean fuzzy analytic hierarchy process (PF-AHP), an expanded method of ordinary fuzzy sets, to evaluate and rank important driver behavior factors built into a hierarchical model using information acquired from drivers’ groups observed in Budapest, Hungary. They claimed that the PF-AHP approach was a helpful method to evaluate driver behavior uncertainty while handling complex road safety issues.

Khademi and Choupani (2018) [23] used the analytic network process (ANP) to analyze the inter-organizational complex system’s behavior to identify major shortcomings in the lead agency of a country for road safety. They applied their developed ANP model on the Road Safety Commission (RSC) of Iran to identify the synthesized influence of factors on each other and suggested a complementary process to specify institutional improvements to prevent any organizational inefficiencies. Yang et al. (2018) [24] devised an analytic network process (ANP) along with statistical methods to evaluate the overall degree of highway safety for Chinese roadways. They analyzed quantitative and qualitative indices (variables), along with factors that influence safety, such as collisions, intersections, alignments, and other important factors. They reported that ANP can take these elements into account, be used to elicit judgments, and generate ratio scales for use in basic arithmetic operations.

Mirmohammadi et al. (2013) [25] studied the prioritization of various elements that can have an impact on the rate and severity of crash issues. They used 15 different safety indicators with a case study in Iran. They applied three different “multiple criteria decision making” (MCDM) methods, viz., analytical hierarchy process (AHP), technique for order of preference by similarity to ideal solution (TOPSIS), and simple additive weighting method (SAW), and reported that AHP performed better compared to others. Farooq et al. (2021) [26] devised a combination of the best–worst method (BWM) and the analytical hierarchy process (AHP) method to compare and quantify the factors for frequent lane changing behavior in relation to road safety. The driver’s replies to a predetermined questionnaire survey were used to prioritize the most important parameters impacting lane changing in order to demonstrate the applicability of the suggested model. They claimed that compared to the traditional AHP, their combined model saw fewer pairwise comparisons (PCs) and more accurate and consistent findings.

Table 1. Comparative summary of references for AHP methods used in traffic safety analysis.

References	AHP Methods	City/Country	Comments
Cheng et al. (2011) [17]	Traditional AHP	China	Some subjective elements in the AHP model were employed; however, they ultimately obtained objective materiality levels of road crashes causes
Fernandeza et al. (2020) [18]	AHP and forced ranking method	Manila, Philippines	Compared AHP with the forced ranking method and reported that the AHP performed better
Saifullizan et al. (2022) [19]	Traditional AHP	Malaysia	Adequacy of the model and technology were key factors in successful application the AHP
Hu et al. (2009) [20]	Improved fuzzy-AHP	China	Assessment index was determined using the scale technique after a summary of the important road traffic safety indices
Farooq et al. (2020) [21]	Fuzzy-AHP (FAHP)	Hungary, Turkey, Pakistan, and China	The FAHP process calculated the weight factors and rated the significant driver behavior criteria built on a three-level hierarchical structure.
Farooq and Moslem (2022) [22]	Pythagorean fuzzy analytic hierarchy process (PF-AHP)	Budapest, Hungary	Evaluated and ranked important driver behavior factors built into a hierarchical model using information acquired from drivers’ groups observed
Khademi and Choupani (2018) [23]	Analytic network process (ANP)	Iran	Identified and synthesized the influence of factors on each other and suggested a complementary process to specify institutional improvements to prevent any organizational inefficiencies
Yang et al. (2018) [24]	Analytic network process (ANP)	China	Analyzed quantitative and qualitative indices (variables) along with factors that influence safety such as collisions, intersections, alignments, and other important factors
Mirmohammadi et al. (2013) [25]	AHP, TOPSIS, and SAW	Iran	Applied AHP, TOPSIS, and SAW methods and reported that AHP performed better compared to others
Farooq et al. (2021) [26]	Best–worst method (BWM) and the analytical hierarchy process (AHP)	Budapest, Hungary	Compared to the traditional AHP, their combined model saw fewer pairwise comparisons (PCs), resulting in more accurate and consistent findings

summarizes references for AHP methods reviewed in this paper.

In

Table 2. Comparison of variants of AHP with the proposed PBAHP method.

Methods	References	Description	Key Features	Comments
Analytical hierarchy process	Cheng et al. (2011) [17], Fernandeza et al. (2020) [18], Saifullizan et al. (2022) [19]	Use pairwise comparisons to prioritize and rank alternatives	Subjective judgments, consistency checks	May suffer from inconsistency in pairwise comparisons Relies on subjective judgments
Fuzzy analytical hierarchy process	Hu et al. (2009) [20], Farooq et al. (2020) [21]	Extension of AHP that incorporates fuzzy logic to handle uncertainties and vagueness	Fuzzy scale on component weight, fuzzy set theory	Requires expertise in fuzzy logic and fuzzy set theory Complex mathematical calculations
Analytic network process	Khademi and Choupani (2018) [23], Yang et al. (2018) [24]	Extension of AHP that models complex decision structures with interdependencies and feedback loops	Incorporates dependence and feedback relationships, super matrix representation	Requires expert knowledge in structuring and modeling the decision problem More complex mathematical calculations Increased computational complexity
Pythagorean fuzzy analytic hierarchy process (PF-AHP)	Farooq and Moslem (2022) [22]	Extension of AHP that incorporates the concept of Pythagorean fuzzy sets and combines fuzzy logic and the AHP	Pythagorean fuzzy subset	Requires expertise in fuzzy logic and fuzzy set theory Complex mathematical calculations Difficulty in obtaining precise linguistic terms
Proportion-based analytical hierarchy process	Current study	Proposed method that uses proportional comparison for pairwise comparisons	Proportional judgments, addresses ratio bias	Addresses ratio bias in AHP Provides more accurate pairwise comparisons Reduces subjectivity in judgments

, several variants of the AHP method used in the literature are compared with the proposed PBAHP. Traditional AHP is one of the methods more frequently used in traffic crash analysis to methodically assess the seriousness and significance of factors involved in traffic crashes. It uses pairwise comparisons and determines the relative importance of various variables, including the state of the roads, driving habits, vehicle features, and environmental factors [17,18,19]. The traditional AHP methods, however, do not consider the actual occurrence or practical impact of each criterion when making a choice, which can produce biased results [27]. FAHP expands AHP by introducing fuzzy logic and helps decision makers deal with the language evaluations and inherent uncertainties that come with traffic crash investigation [20,21]. The FAHP enables a more flexible and thorough assessment of crash variables by utilizing fuzzy numbers and logic to better capture the subjective assessments. Another extension is P-FAHP, which uses Pythagorean Fuzzy subsets, considering the degree of membership and nonmembership of factors [22]. The ANP method adopts a more comprehensive approach by taking into account both the interdependencies and the hierarchy of crash reasons. ANP captures the interconnectedness of crash causes, offering a comprehensive view of crash analysis and facilitating a more in-depth comprehension of system dynamics. Despite the advantages of the AHP’s multi-criteria decision-making technique, some limitations are frequently present [28]. In this present study, proportion-based or percentage-based AHP (PBAHP) is proposed, which is a different approach that considers the relative importance of each criterion in the decision-making process and has been offered as a solution to this problem. PBAHP can be promising as a method for identifying the main causes and types of major crashes. More details about this technique are provided in the proceeding sections of this paper.

Overview of the Analytical Hierarchy Process (AHP) and the Proportion-Based Analytical Hierarchy Process (PBAHP)

AHP is a multicriteria decision-making tool which has been used in a wide range of applications. These applications include social and political sciences, as well as engineering processes [29]. The literature shows that AHP was first proposed by T. Saaty as early as 1972. Since then, the process has adopted many developments, enabling its greater use in different fields. These include the establishment of ranking scales, inconsistency limits and indices, and the development of easy-to-use exclusive software packages for AHP [30].

AHP is a measurement tool which works on pairwise comparisons based on fixed scales. The measuring scale can use discrete ratings of actual measurement of user preferences [31]. The problem is divided into two levels; the first level is the “criteria”, and the second level is called the “goal”. Criteria refers to the factors that affect the choice or the outcome, while goal is the choice or outcome of the process. There are six steps to formulating AHP methods [32,33,34]:

• First step: problem identification and set up of evaluation standards;
• Second step: setting up the AHP hierarchy;
• Third step: creating matrices for pairwise comparison of the criteria;
• Fourth step: comparison matrices’ normalizing;
• Fifth step: calculation of the priority vectors;
• Sixth step: calculation of consistency ratio (CR).

Following the abovementioned steps for the pairwise comparison of the criteria, their rankings from each comparison are averaged out and normalized to obtain the relative weight of the criteria, as per Equation (1).

W = [(∏_i=1,n R_i)^1/n]/[∑_i=1,n (∏_i=1,n R_i)^1/n]

(1)

where

• R = ranking for each pair of factors;
• n = number of pairs of factors.

The pairwise comparison was performed for each goal based on one criterion at a time. The rankings from multiple comparisons of goals were combined, similar to criteria, to obtain their weights. The resulting weight of each goal is the dot product of its weight, calculated for each criterion, and the weight of the specific criterion.

As the weights (as explained above) are calculated using multiple independent comparisons, there is a chance of inconsistency in the rankings. To consider that issue, the consistency ratio (CR) and consistency index (CI) are calculated for the matrix resulting from multiple comparison at each stage. CR is a measure of variation within the data, derived from the eigen vector of the ranking matrix, while CI is a relative ratio for the CR with the standard randomness index given by Saaty (2005) [35]. These values can be determined using Equations (2) and (3).

CR = (λ − n)/(n − 1)

(2)

CI = CR/RI

(3)

where

• λ = maximum value of eigen vector for the ranking matrix;
• n = size of the matrix;
• RI = random consistency index, given by [35].

The rankings of criteria and goals are normally taken using the perception or opinion of respondents to a survey [36].

In AHP, a complicated problem requiring decision making is divided into smaller, more manageable components, and each component is assessed in relation to the others. According to their relative importance, the criteria and options are compared pairwise in the conventional AHP.

This research proposes a new proportion-based approach for calculating ratings for each event and crash severity. It is a newly developed variant of AHP that evaluates the components using percentage scores of the requirements that each alternative meets, rather than using ratio scales. PBAHP’s capability to consider the comparative relevance/occurrence of each element causing traffic crashes is one of its primary features. PBAHP assesses each factor’s significance using percentage scores, enabling a more precise evaluation of the elements that contribute to serious collisions. Furthermore, PBAHP is a straightforward approach that is simple to comprehend and use. It is expected that the current methodology can be adopted in a variety of fields for the application of AHP, wherein the available data are extracted from an experiment or field. To the best of the authors’ knowledge, proportion-based ratings have not been used for AHP until now.

3. Study Area, Data Sources, and Data Description

Al-Ahsa is located in the Eastern Province of Saudi Arabia and is bordered by the Persian Gulf to the east. The area extends from 29°20′ N in Kuwait to the southernmost point of the Gulf of Bahrain at 25°10′ N (Figure 1). The area is well known for its agricultural output, historic buildings, and palm trees. It is one of Saudi Arabia’s largest regions, and has a vibrant culture and history. The area is also home to several historical attractions, such as the Qara Mountain Caves, which are thought to be more than 7000 years old and are home to prehistoric rock art. UNESCO has recently listed it as a heritage site in Saudi Arabia. The Al-Ahsa region was selected for this study since previous research has shown a high rate of traffic crashes in Al-Ahsa, with 31.9% of all crashes in the Eastern Province occurring in this city between 2009 and 2016. Dammam, Hafr Al-Batin, Qatif, Jubail, Dhahran, and Khobar are the next cities in terms of the number of crashes, with crash rates below 5% in other cities (Figure 2) [37]. These findings indicate that Al-Ahsa is particularly susceptible to traffic crashes in the Eastern Province. Although Al-Ahsa possesses significant potential to thrive as an international tourist attraction because of its recognition as a UNESCO-listed heritage site and a Guinness World Record, its high crash rate may put this at risk, necessitating immediate action. During the period of October 2014 to May 2018, the Road Safety Authority in Dammam provided data on traffic crashes in Al-Ahsa. The data revealed that around 8% of fatal and 30% of serious injury incidents were due to crashes between cars. Fatal crashes refer to those which involve at least one death, while serious injury crashes refer to those crashes in which at least one of the victims is hospitalized, without involving any deaths. The data provided by the Road Safety Authority also included the classification of crashes on the bases mentioned above. Another significant type of crash was vehicle overturning, which contributed to 6.5% of fatal and injury events. Pedestrians were also shown to be vulnerable, as they were involved in 12.5% of injury incidents and 2% of fatal crashes. Although some types of crashes, such as crashes with road guardrails, parked vehicles, motorcyclists, and permanent objects, have low fatality rates, they still pose a risk of harm, as illustrated in Figure 3. Therefore, further analysis and actions are needed to address the issue of traffic crashes in Al-Ahsa.

Data Description

Table 3. Distribution of crash types and reasons in the dataset.

Crash Types	Number of Crashes	Frequency of Crashes	Crash Reasons	Number of Crashes	Frequency of Crashes
Collision	1542	38.54%	Sudden lane changing	1918	47.91%
Vehicle overturned	862	21.54%	Speeding	725	18.11%
Hit pedestrian	582	14.55%	Not giving way	615	15.36%
Hit road fence	278	6.95%	Insufficient safe distance	256	6.40%
Hit motorcycle	201	5.02%	Driver distraction	230	5.75%
Hit parked vehicle	154	3.85%	Crossing without pedestrian crossing	39	0.97%
Hit electric post	100	2.50%	Illegal overtaking	22	0.55%
Hit animal	61	1.52%	Red light violation	13	0.32%
Hit bicycle	53	1.32%	Driving opposite to traffic	10	0.25%
Hit roadside barrier	52	1.30%	Not stopping at stop sign	7	0.17%
Undefined category	37	0.92%	Drifting	5	0.12%
Hit fixed object	34	0.85%	Falling asleep	2	0.05%
Hit tree	25	0.62%	Getting out of moving vehicle	2	0.05%
Fell off the slope	7	0.17%	Hanging on the outside of vehicle	2	0.05%
Hit plate	4	0.10%	Unsafe road works	2	0.05%
Fell off bridge	4	0.10%	Exhaustion	1	0.02%
Fire on vehicle	3	0.07%	Violating pedestrian sign	1	0.02%
Hit signal	2	0.05%	Downhill	1	0.02%
Hit waste container	0	0.00%	No warning signs	1	0.02%

shows the number of crashes reported for each crash type, and the reason. Crash type refers to the type of impact which the vehicle had during the crash, while crash reason refers to the cause of the crash leading up to it. The first step was to merge the types and reasons that had insignificant reported crashes into a category of “others”. The limit of significance was set as 5%, and therefore all those crash reasons and types with less than 5% crashes were merged in the above category. This limit was set as per the normal practice of statisticians in which the tests for significance were set at the 5% confidence level [38]. The crash types and reasons which had a proportion above 5% were considered individually, and they are highlighted in Table 3, while the rest were merged into the “others” category for each case. They are also shown in Figure 4 and Figure 5.

4. Modeling Methodology

The first step of the AHP is to structure the problem with respect to its criteria and goals. For this research, the criteria were the crash types and reasons, while the goal was the crash severity. The structure can be observed from Figure 4 and Figure 5; the process was applied exclusively for the crash type and then for crash reason. The ranking for each criterion and goal were calculated, as shown by Equations (4) and (5), while the total weight for each crash severity was calculated as per Equation (6).

As stated earlier, this research employed the PBAHP method for the analysis. The difference between this method and traditional AHP is the use of proportions for calculating the ranking of the factors (crash types and reasons, in this case), as shown in Equations (4) and (5). The traditional AHP method uses a survey or interview and asks the respondents to carry out a pairwise comparison [39]. To achieve this, the respondent hypothetically generates the ratio of importance of each factor while answering the comparison questions. The same approach can be applied to quantitative data, such as crash data, by applying a method to evaluate the relative importance of one event over the other, which can be used in place of the ratings given by respondents in the traditional AHP method. The present study used a simple and efficient method to evaluate the relative importance of each crash event (type or reason) using the proportion of its occurrence in the crash data. This method makes use of the available data without the risk of receiving biased responses and does not require complicated or lengthy preprocessing for using the data. In the present research, these rankings were used to determine the relative weights (based upon their occurrence) for the crash severity. Higher rankings/ratios indicated a higher proportion of certain crash types or reasons in comparison to the other, which shows their comparatively higher importance or significance for crash severity. For example, if the ratio is 3 between crash type A and B, then this indicates that type A has 3 times more importance (weightage) than type B, and vice versa, in the available dataset. This is further explained in Section 5.

Rij = Pi/Pj

(4)

where

• Rij = Ranking of reason/type ‘i’ in comparison to “j”;
• Pi = Proportion of crashes by reason/type “i”;
• Pj = Proportion of crashes by reason/type “j”.

Rxyi = Pxi/Pyi

(5)

where

• Rxyi = Ranking of crash severity “x” over “y” based on crash type/reason “i”;
• Pxi = Proportion of crashes belonging to severity “x” out of those caused by crash type/reason “i”;
• Pyi = Proportion of crashes belonging to severity “y” out of those caused by crash type/reason “i”.

The total weight for each crash severity was the weighted sum of probability of each crash severity based on the contributing factor, as shown in Equation (6). This was used to develop the utility function for the crash severities.

$$\mathrm{TW}(\mathrm{X})={{\displaystyle \sum }}_{\mathrm{i}=1}^6{\mathrm{W}}_{\mathrm{xi}}{\mathrm{W}}_{\mathrm{i}}$$

(6)

where

• TW = total weight for crash severity “X”;
• Wxi = weight of severity “X” with crash type/reason “i”

  Table 4. Ranking and relative weights for crash types.

  Crash Types	Collision	Hit Motorcycle	Hit Road Fence	Hit Pedestrian	Vehicle Overturned	Others	Relative Weights
  Collision	1.00	7.77	5.54	2.65	1.79	2.88	1.00
  Hit motorcycle	0.13	1.00	0.72	0.34	0.23	0.37	0.13
  Hit road fence	0.18	1.39	1.00	0.48	0.32	0.52	0.18
  Hit pedestrian	0.38	2.94	2.08	1.00	0.68	1.08	0.38
  Vehicle overturned	0.56	4.35	3.13	1.47	1.00	1.61	0.56
  Others	1.00	7.77	5.54	2.65	1.79	2.88	1.00
  CI							0.00

  or

  Table 5. Ranking and relative weights for crash reasons.

  Crash Reasons	Driver Distraction	Speeding	Not Giving Way	Sudden Turning	Insufficient Safe Distance	Others	Relative Weights
  Driver distraction	1.00	0.32	0.37	0.12	0.90	0.89	0.06
  Speeding	3.13	1.00	1.18	0.38	2.83	2.80	0.18
  Not giving way	2.70	0.85	1.00	0.32	2.40	2.37	0.15
  Sudden lane changing	8.33	2.63	3.13	1.00	7.49	7.40	0.48
  Insufficient safe distance	1.11	0.35	0.42	0.13	1.00	0.99	0.06
  Others	1.12	0.36	0.42	0.14	1.01	1.00	0.06
  CI							0.00

  ;
• Wi = weight of crash type/reason “i”, from Table 5 or

  Table 6. Ranking and relative weights for the severity of crashes according to crash type.

  Crash Severity	Fatal	Serious Injury	Weights
  Collision
  Fatal	1	0.25	0.33
  Serious injury	4	1	0.67
  Hit motorcycle
  Fatal	1	0.09	0.23
  Serious injury	11.11	1	0.77
  Hit road fence
  Fatal	1	0.1	0.24
  Serious injury	10	1	0.76
  Hit pedestrian
  Fatal	1	0.16	0.28
  Serious injury	6.25	1	0.71
  Vehicle Overturned
  Fatal	1	0.43	0.4
  Serious injury	2.33	1	0.6
  Others
  Fatal	1	0.14	0.27
  Serious injury	7.14	1	0.73

  .

5. Results and Discussions

Figure 6 shows the distribution of crashes based on type, while Figure 7 shows the same on the basis of crash reasons. Using Equation (4), rankings were calculated for each crash type and reason, and are presented in Table 4 and Table 5. These tables also show the weights of each criterion and the CI for each matrix, calculated as per Equations (2) and (3).

Table 4 shows that “collision” was the highest-contributing type of crash to fatal and serious injury crashes in the study, followed by “vehicle overturned” and “hit pedestrian”. Figure 6 also points to the same fact.

Inconsistency was “0”, as the rankings were derived from the ratios of the crash proportions. This parameter can be an issue when rankings are made by the respondents based on their judgements [40].

Table 5 shows that “sudden lane changing” was the highest contributing reason for fatal and serious injury crashes, followed by the vehicle overturning. This finding was also corroborated by the fact that the most common type of crash was the “collision”, which is bound to happen during the sudden turning of a vehicle. Other major contributing reasons to fatal and serious injury crashes were speeding and not giving way. The trends are also evident in Figure 7.

Inconsistency for these weights was also found to be 0, for the reasons explained above for the crash types.

Figure 8 and Figure 9 show the distribution of crash severities based on crash types and reasons, respectively. The rankings of each severity on the basis of each criterion were calculated as per Equation (5). These rankings are shown in Table 6 and

Table 7. Ranking and relative weights for the severity of crashes according to crash reasons.

Crash Severity	Fatal	Serious Injury	Weights
Driver distraction
Fatal	1	0.14	0.27
Serious injury	7.14	1	0.73
Speeding
Fatal	1	0.23	0.32
Serious injury	4.35	1	0.68
Not giving way
Fatal	1	0.15	0.28
Serious injury	6.67	1	0.72
Sudden lane changing
Fatal	1	0.26	0.34
Serious injury	3.85	1	0.66
Insufficient safe distance
Fatal	1	0.22	0.32
Serious injury	4.54	1	0.68
Others
Fatal	1	0.36	0.38
Serious injury	2.78	1	0.62

. The weights for each crash severity based on each criterion are also shown in these tables.

The inconsistency check was not required since there were only two outcomes to be compared. The chances of inconsistency are produced through multiple pairwise comparisons. This is the reason that the inconsistency ratios (IRs) for 1 and 2 dimensional matrices are given as “0” [35].

Fatal crashes had the highest weight among “vehicle overturned” crashes, while the highest for serious injury crashes was when “hit motorcycle” occurred. The crash type “collision” also had a relatively high weight for fatal crashes. It should be noted that a higher weight for a certain severity level reduces the weight for the others, as the total of the weights should be “1”. It seems that the chances of fatalities in crashes increase when a vehicle is overturned, which could be due to the loss of control by the driver. These observations are also confirmed in Figure 8. The total weights were calculated as per Equation (6) for different crash severities, and they are given in Table 6 based on different crash types and reasons.

Inconsistency calculations were not required, as explained earlier for crash types. The fatal crashes had the highest weight for “sudden lane changing” crashes, while the highest weight for serious injury crash was from “driver distraction”, and “leaving insufficient safe distance” and “speeding” also caused similar and relatively high weights for fatal crashes. These trends were corroborated by Figure 9. Hence, it can be said that fatal crashes are more likely to occur when drivers are not following road regulations properly or do not have enough time to respond to events.

It can be observed that the overall weight for each severity type remained the same whether they are taken on the basis of crash types or reasons as shown in

Table 8. Total weights for each crash severity type.

Contributing Factor	Crash Severity
	Fatal	Serious Injury
Crash Types	0.322	0.686
Crash Reasons	0.321	0.669

. This was expected, since it would have related to their overall proportion in the available sample. Figure 10 and Figure 11 show the combined weights for the severity of the crash based on each crash type and reason.

Figure 10 shows that collisions and vehicles overturning are the two most common types of crashes which cause fatality or serious injury in a crash. These weights were calculated as the cross product of the relative weights of a factor (provided in Table 4 for crash type and Table 5 for crash reason) and the relative weight of factor with the severity level (Table 6 or Table 7). These weights indicate the relative chance of a severity type, among the entire dataset of crashes, with a particular type or reason for the crash. The weight or relative importance of “collision” to cause a serious injury crash was at least 10% higher than any other crash type. Similarly, it also had the highest weight for fatal crashes as well; however, the difference in its weight (approximately 5%), compared to other types, was not so prominent for fatal crashes as it was for serious injury crashes. Motorcycle riders were proven to be vulnerable road users because of their greater involvement in crashes of higher severity [41].

Figure 11 shows that sudden lane changing was the most common reason for serious injury and fatal crashes, followed by speeding. The importance of “sudden turning” was at least 20% higher than any crash reason for serious injury crashes, while it was at least 10% higher than any other crash reason for fatal crashes. The least common reason for serious injury and fatal crashes was driver distraction. This could be because distracted drivers tend to become slower and keep a higher headway between the vehicles, as has been proven by other studies [42]. The combined weights could be used to form a utility function for the prediction of crashes, as proposed by [43]. Hence, the utility function for different crash severities can be listed as Equations (7) and (8). The development of such utility functions would have been unlikely with traditional statistical approaches for the type of data available, because of their statistical constraints.

U_Fatal = 0.129 (Collision) + 0.012 (Hitmotorcycle) + 0.017 (Hitroadfence) + 0.042 (Hitpedestrian) + 0.088 (Vehicle overturned) + 0.035 (Otherstypes) + 0.016 (Driverdistraction) + 0.058 (Speeding) + 0.042 (Notgivingway) + 0.163 (Suddenlanechanging) + 0.019 (Insufficientsafedistance) + 0.023 (Otherstypes)

(7)

U_{Seriousinjury} = 0.261 (Collision) + 0.038 (Hitmotorcycle) + 0.053(Hitroadfence) + 0.106 (Hitpedestrian) + 0.132 (Vehicleoverturned) + 0.095 (Othertypes) + 0.044 (Driverdistraction) + 0.122 (Speeding) + 0.108 (Notgivingway) + 0.317 (Suddenlanechanging) + 0.041 (Insufficientsafedistance) + 0.037 (Otherreasons)

(8)

6. Conclusions

In Saudi Arabia, traffic crashes are a major public health hazard because of their various causes, which include aspects relating to the driver, the road, and the vehicle. Understanding the root causes of traffic crashes in the country is crucial for developing effective strategies to stop these occurrences and improve road safety. Understanding the typical reasons and crash types that lead to severe crashes is a crucial first step in developing effective solutions for severe crash prevention. In this research, the PBAHP technique was used, and its potential in discovering the primary causes and crash types that result in catastrophic crashes was proven. The PBAHP was used to identify and rank the causes of traffic crashes, as well as to assess the efficiency of safety measures.

According to this study, “collision” crashes were the most common cause of fatalities and serious injuries, followed by “vehicle overturned” and “hit pedestrian”. This conclusion is further supported by the fact that “collisions”, the most frequent sort of crash, are invariably caused by rapid turns of a vehicle. Speeding and failure to yield are two other significant causes of death and serious injury collisions. Using the same process, the combined effects of events on the crash severity (i.e., the interaction of different crash types and reasons) could be analyzed. However, the limitation of this study is that it could not be performed with the present dataset because of the lack of samples for each combination. Thus, future studies should be based on enhanced datasets. As an alternative, the total weights from the PBAHP process were used to develop utility functions for fatal and serious injury crashes. These functions could be used for estimating the likelihood of crashes in different scenarios.