Auto-Oriented Communities in Developing Countries: Bus Rapid Transit Implementation Prospects

Abstract

Public transportation services often face challenges in middle- and low-income nations from both public acceptance and economic constraints. Jordan is classified as a middle-income country with a population of over 10 million, 4 million of whom live in Amman, the capital. The development and operation of a bus rapid transit (BRT) system in Amman city was recently proposed. The BRT project is anticipated to offer a solution to the city’s escalating congestion problem. This study’s objective is to conduct a “before” analysis to identify the variables that affect the willingness of people to use the Amman BRT system. The socioeconomic characteristics and travel habits of individuals were used to model the willingness to use BRT. An online survey was distributed to Amman residents and 238 valid responses were returned. Two popular techniques were utilized: binary logistic regression and Bayesian networks. Ten models were developed: one binary logistic model and nine Bayesian network models. The results of these models were compared based on accuracy, sensitivity, specificity, area under the Receiver Operating Characteristic (ROC) curve, complexity, and number of selected variables. It was found that Bayesian networks were more effective in modeling willingness to use BRT. Willingness to use BRT was shown to be higher among households without cars, youths, females, and university students, and if there were fewer transfers along the route. It became clear that introducing a new public transportation system is well appreciated, particularly in areas with low income, insufficient existing public transportation services, and where driving a car is the norm.

Keywords

public transportation bus transit systems bus rapid transit capacity and quality of service customer satisfaction/loyalty quality

Bus rapid transit (BRT) systems, which have been in use for more than 40 years worldwide, are known for being more affordable than other rapid transit systems like metro and light rail transit. Numerous studies have examined how BRT users perceive the service and their preference to travel in metropolitan areas. According to research on public attitudes toward BRT services and their willingness to live near to BRT stations, BRT is favored by users and enhances the attractiveness of residential properties ( 1 ).

BRT is always positioned in the middle between urban rail and conventional bus systems when compared with other public transportation systems, in particular on quality and cost. Because BRT combines the operating flexibility and reduced cost of the conventional bus system with the reliability and speed of rail, it has been referred to as “the best of both worlds.” BRT systems have the potential to have substantial effects on urban economic, social, and environmental development, just like other types of public transportation ( 2 ). Similarly, the BRT project was proposed by Amman city planners as part of a larger strategy to change the city’s social and urban structure ( 3 ).

Jordan is considered an upper lower-income country ( 4 ). More than 42% of Jordan’s population live in Amman where most of the businesses, universities, and health services are available. However, most of the time, it is not possible to find accessible, efficient, safe, economical, and sustainable public transportation options in the city. The existing modes of transportation offered to Amman residents and workers lack schedules, safety, and occasionally security.

According to data from the Land Transport Regulatory Commission, the body in charge of supervising Jordan’s public transportation services, there were 37,184 vehicles in Jordan’s entire public transportation fleet at the end of 2020. These included taxis and white cabs (shared taxi), as well as buses with a capacity of up to 50 passengers and minibuses with a capacity of up to 23 passengers ( 5 ).

With a population of 11.057 million people ( 6 ), Jordan has seen an increase in transportation demand during the past 15 years, since 2008. However, the necessary development of high-quality public transportation has not kept up with demand. Approximately 13% of trips are made by public transportation, which is a very low percentage. Around 33% of trips in Jordan are made in private automobiles, and between 2008 and 2018, the number of private vehicles nearly doubled, reaching over one million.

During the same time period, the number of public vehicles only increased by 45%, from over 29,000 to over 42,000 ( 7 ). As for the relationship between income and car ownership, developing countries generally follow a pattern quite similar to that of industrialized countries, with fewer than one car per 10 people in most developing countries, compared with four or more cars per 10 people in wealthy industrialized countries ( 8 ). In Jordan, it was estimated that there were 0.58 buses per 1,000 inhabitants in the fourth quarter of 2021 ( 9 ).

A prospective concept to build a BRT system was put forth in 2008 in an effort to address the current challenges and issues with the existing public transportation systems, understanding the economic situation of Amman inhabitants. In Amman, the average household annual income is 12,657 Jordanian Dinars (JOD) (17,827 USD), according to the most recent national survey, which was carried out in 2017/2018 ( 10 ). As a percentage of overall spending, the average household spends roughly 16% on transportation, second among all annual household expenses ( 6 ). The BRT project was delayed until 2021 by a lack of funding and other administrative obstacles. The ongoing project schemes, according to sources from the Amman Greater Municipality, call for a planned time headway of no more than 3 min from and to contemporary, fully integrated bus terminals ( 11 ).

Given its novelty, there is a scarcity of research that has assessed and analyzed the Amman BRT. The only two studies that examined certain elements of the Amman BRT were concerned with public involvement ( 12 ) and sustainability ( 13 ).

Al-Sharari ( 12 ) sought to assess the public’s participation in the development and implementation of the Amman BRT project. In that study, a thorough review of relevant literature and current policies, semi-structured interviews with significant project stakeholders, and an online poll aimed at the general public were all carried out. The Amman BRT public participation strategy was judged to be insufficient and to fall short of addressing the people’s concerns about the project.

To evaluate the Amman BRT sustainability, Al-Dalain and Alnsour ( 13 ) looked into the sustainability performance metrics in relation to the economic, environmental, and social dimensions. The literature review served as the initial instrument for producing indicators, and semi-structured interviews with five subject-matter experts were then conducted. According to the results, there are 13 factors that can be used to evaluate how well the Amman BRT system performs on sustainability. Additionally, the findings revealed that while the overall indicators are relevant, the economic dimension of affordability is the most crucial.

To that purpose, based on the current state of the available public transportation services, this research aims to investigate the factors that are expected to contribute to people’s willingness to use the BRT system.

Literature Review

Different methodologies were employed in previous studies to evaluate the level of service provided by public transportation. Traditional generalized linear models (GLM), structural equation models (SEM), and data mining techniques have all been utilized in the literature to investigate and model the quality of service provided by public transportation.

Many researchers employed GLMs to model the level of service for public transportation. Stojic et al. ( 14 ) used survey data to model the consumer satisfaction of younger commuters using regression techniques. Only four of the 13 independent variables—general travel information, general onboard comfort, general bus stop condition, and punctuality/price—were discovered to have an effect on overall satisfaction. To model the quality of service provided by public transportation systems, other researchers have also employed regression analysis techniques such logit or probit ( 15 – 18 ). However, these methods are all affected by presumptions about the distribution to which the variables belong, which may lead to inaccurate conclusions ( 19 ).

Other researchers employed SEM to model service quality. Nguyen-Phuoc et al. ( 20 ) studied the riders’ loyalty as determined by ride-sourcing services using SEM. The model’s input data were obtained via a questionnaire, and SEM findings showed that satisfaction, perceived service quality, and involvement can all be used to measure passenger loyalty. Perceived safety, however, did not perform well as a predictor of loyalty.

De Oña et al. ( 21 ) investigated how passengers perceived the quality of the service, gathering data through a questionnaire and modeling with SEM. The findings showed that the factors that most significantly influenced perceived service quality were frequency, timeliness, speed, closeness, and fare.

SEM technique tests the fit of a proposed “causal” model under the assumption that equations are linear. Model convergence may be a challenge because SEM involves numerous pathways and many variables. Additionally, modeling using categorical variables is carried out under the presumption that a quantitative multivariate normal variable exists underneath that belongs to a specific category if it falls within a specific range ( 22 ).

Data mining techniques were also utilized to model the quality of public transportation services in an effort to create models devoid of assumptions about the distributions to which the data in question belongs. Several researchers have used decision trees (DTs) (e.g., 19 , 23 –26). De Oña et al. ( 19 ) used DTs and clustering, with the latter being used to categorize passengers into homogeneous groups. DTs were used to examine the service quality and identify the factors that affect it as a whole. Frequency, punctuality, speed, safety, and space were found to be dependent on the group of passengers analyzed for the entire sample of users.

Artificial neural networks (ANN) have also been found to be employed in the analysis of public transportation systems. The quality of public transportation services as perceived by the passengers was examined by Garrido et al. ( 27 ) using ANN. In their work, they compared three ANN algorithms, and the findings showed that frequency was the most key factor determining the quality of service, followed by speed, information, and proximity. Unlike other methods, like DTs ( 28 ) or Bayesian networks (BNs) ( 29 ), however, ANN lacks the simplicity and graphical representation that these methods provide.

BNs have been successfully utilized to assess traffic accidents ( 29 – 34 ). When modeling the relationships between personal characteristics and satisfaction, Perucca and Salini ( 35 ) set the bar for the use of BN in the study of customer surveys of railway systems. They uncovered evidence that BN has a stronger predictive power than the popular ordered logistic regression. In evaluating the performance of public transportation, Wu et al. ( 36 ) used BNs and revealed causal relationships between service-related factors. They have demonstrated the benefits of BNs in the analysis of transit service.

Wu et al. ( 37 ) used BNs to analyze survey data from Nanjing, China to determine which public transportation characteristics have the greatest effects on customer satisfaction. They also looked into how these characteristics are correlated with one another. The resulting BN tree was used to depict the correlations between service characteristics and overall passenger satisfaction.

Mandhani et al. ( 38 ) used an integrated methodological approach to enable the systematic identification of hidden relationships among service quality factors in the metro rail transit system of Delhi, India. On the perceptions of 2,390 Delhi Metro riders, the study used an integrated BN and partial least squares SEM (PLS-SEM) technique. First, using principal component analysis, the study divided its 41 service quality indicators into eight service quality factors. Second, to develop the most reliable network structure, the extracted factors were trained in BN. Third, a service quality model was developed using PLS-SEM to test and analyze the robust BN structure.

Díez-Mesa et al. ( 39 ) employed BNs in a two-step approach to simulate the level of service in the metropolitan light rail transit service of Seville, Spain. The two-step procedure incorporated SEM and BNs. Without making any presumptions, they used their approach to determine the correlations between the service’s attributes and the passengers’ overall assessment of the service’s quality.

To the authors’ knowledge, however, BNs have never been used to assess people’s willingness to take public transportation. BNs can avoid many of the problems that many of the modeling methods outlined earlier encountered. The challenges of having prior knowledge of the variables employed are overcome by BNs, which learn directly from data without presumption ( 39 ).

To this purpose, the results of the BNs analysis of willingness to use BRT services will be compared with those from logistic regression, the most popular regression method used to evaluate public transportation services.

Methodology

This section describes the methodology used in this study. First, BN and binomial logistic regression (BLR) techniques are defined, followed by a detailed analysis of the methodological process, and finally the sample design is presented.

Bayesian Networks

BN is a directed acyclic graph over a set of variables (U) and a set of probability tables (Bp). A BN represents joint probability distributions which are calculated based on the following Equations 1–3:

P (U) = \underset{xi ϵ U}{Π} p (xi | pa (xi))

(1)

U = (x 1; x 2; \dots; xn)

(2)

B_{p} = p (xi / pa (xi))

(3)

where n ≥ 1, xi∈U, pa(xi) is the set of parents of xi in BN and i = (1; 2; 3; …; n).

According to the BN theory, BNs are graphical models of interactions among a set of variables, where the variables are represented as nodes (vertices) and the relations between variables (interactions) are represented as the arcs (edges) ( 40 ). It is worth mentioning that the BN can satisfy the probability distribution of the variables without the arcs being causal since the arcs between variables in a non-causal BN could indicate a sort of interrelationship(s) among these variables. In the classification process the classifying variable, y = x₀, is called the class variable, given a set of variables U which is also called attribute variables. A classifier, h: U → y, is a function that classifies an instance of U to a value of y. A data set, D, containing samples over (U, y) is used to learn a classifier in which the learning task consists of finding an appropriate BN given a data set D over U.

The learning of both structure and variables is performed when there are significant amounts of data, and it is necessary to evaluate them and develop a model for predicting the behavior of unobserved cases ( 41 ). The two primary methods for structure learning in BNs are as follows:

Constraint-based: Run tests of conditional independence on the data, and search for a network that fits the observed dependencies and independencies.

Score-based: Establish a score that evaluates how well a structure’s dependencies or independencies match the data, then search for a structure that maximizes the score.

The merit of score-based methods over constraint-based methods is that they are less vulnerable to errors in individual tests; trade-offs may be made between how dependent the variables are on one another in the data and the penalty of adding the edge (arc). The score-based method is used in this study because of the advantages previously mentioned.

In this research work, hillclimbing, K2, and simulated annealing search algorithms and Bdeu, minimum description length (MDL), and Akaike information criterion (AIC) scores were applied since they are widely used, relatively quick, and produce good results in overall network performance and complexity ( 32 , 42 ). A search heuristic is used to suggest option networks, and a scoring function is used to determine which of any two options is more likely given the training data for every given pair of candidates. The following is a definition for each search algorithm and score metric used.

Hillclimbing search algorithm is an optimization method for tackling challenging computational issues. Through the use of an evaluation function, hillclimbing makes an attempt to incrementally improve the current state. Progress in hillclimbing is only possible if there is a higher point in the surrounding landscape ( 43 ).

A K2 search algorithm starts off presuming that a node has no parents, and then it gradually adds the parent whose inclusion will most likely result in a network. Until no further parents may raise the likelihood of a structure, parents are greedily added to a node. The process is performed for each node in the order indicated by the node ordering ( 41 ).

Simulated annealing search algorithm is a stochastic method for finding a valid approximation to a function’s global minimum. It creates random points near the present best point and assesses the problem functions there. The point is accepted, and the best function value is changed if the cost function value is lower than its existing best value ( 44 ).

BDeu score metric, MDL, and the AIC score functions were used to develop BN structures.

Let r_i (1 ≤ i ≤ n) represent the cardinality of x_i, q_i serves to indicate the cardinality of the parent set of x_i in BN, that is, at how many different values x_i’s parents can be instantiated. Consequently, q_i can be computed as the product of the cardinalities of the nodes in pa (x_i ),

$q_{i} = \underset{x_{j} \in pa (x_{i})}{Π} r_{j} . Noting that pa (xi) = ϕ implies qi = 1 .$

denotes the number of records in D for which pa(x_i) takes its j ^th value. $N_{ijk} (\leq i \leq n, 1 \leq j \leq q_{i,}, 1 \leq k \leq r_{i})$ denotes the number of records in D for which pa(x_i) takes its j^th values and for which x_i takes its kth value. So, $N_{ij} = \sum_{k = 1}^{r_{i}} N_{ijk} .$ N denotes the number of records in D.

Let the entropy metric H (BN, D) of a network structure and database be defined as in Equation 4:

H (BN, D) = - N \sum_{i = 1}^{n} \sum_{j = 1}^{q_{i}} \sum_{k = 1}^{r_{i}} \frac{N_{ijk}}{N} \log \frac{N_{ijk}}{N_{ij}}

(4)

and the number of parameters K is defined as in Equation 5:

K = \sum_{i = 1}^{n} (r_{i} - 1) . q_{i}

(5)

The AIC metric is defined as in Equation 6:

Q_{AIC} (BN, D) = H (BN, D) + K

(6)

A term P(BN) can be added representing prior information over network structures, but will be ignored for simplicity in the WEKA implementation ( 45 ).

The MDL metric Q_MDL(BN,D) of a BN structure for a database D is defined as in Equation 7:

Q_{MDL} (BN, D) = H (BN, D) + \frac{K}{2} logN

(7)

The BDeu metric is defined as in Equation 8:

Q_{BDe} (BN, D) = P (BN) Π_{i = 0}^{n} Π_{j = 1}^{q_{i}} \frac{Γ (1 / q_{i})}{Γ ((1 / q_{i}) + N_{ij})} Π_{k = 1}^{r_{i}} \frac{Γ ((1 / r_{i}) . q_{i} + N_{ijk})}{Γ ((1 / r_{i}) . q_{i})}

(8)

where P(BN) is the prior on the network structure (taken to be constant, so it is ignored in WEKA implementation) and Γ(-) is the gamma function ( 45 ).

BNs are used to create various models and assess their ability to classify willingness to use BRT, here divided into either “yes” or “no.” The correct classification means that the developed models are capable of distinguishing between “yes” and “no” without a high percentage of misclassifications. Misclassifications indicate that the model classifies “yes” as “no” or vice versa. In a two-class problem, a classifier is generally said to have a high percentage of misclassifications if it incorrectly classified more than 50% of the data, with 50% representing the level of chance that would be expected from a random guess ( 46 ). The misclassifications are measured using several performance measures that will be described in the “Results and Discussion” section.

Binary Logistic Regression

BLR is one type of generalized linear model that can investigate the dependency of a binary response on a set of explanatory variables. These explanatory variables include discrete, continuous, or a combination. BLR is adopted in this study since willingness to use BRT (response variable) is a binary variable and the independent variables are categorical. For survey respondents, the explanatory variables are represented by X = (X1, X2, …, Xk), the observed value is represented by xi = (xi1, xi2, …,xik), and the binary response variable is Y where Yi = 1 if the respondent rides a bus and Yi = 0 if other ( 47 ). The probability (π) that the person i is a bus rider can be calculated using the following Equations 9 to 12:

π_{i} = \Pr (Y_{i} = 1 | X_{i} = x_{i}) = \frac{\exp (β_{0} + β_{i} x_{i})}{1 + \exp (β_{0} + β_{i} x_{i})}

(9)

logit (π_{i}) = \log (\frac{π_{i}}{1 - π_{i}}) x_{ik}

(10)

= β_{0} + β_{i} x_{i} x_{ik}

(11)

= β_{0} + β_{1} x_{i 1} + \dots + β_{k} x_{ik}

(12)

Methodological Process

To illustrate the procedure followed in this study, Figure 1 illustrates the methodological procedure employed.

Figure 1.

Methodological process.

As shown in Figure 1, the data set obtained from the survey was split into 70% for training set and 30% for test set. It is important to split the data into the training set and the testing set while learning relationship(s) from the data to prevent overfitting. Data from the testing set are used to determine the model’s accuracy after the model has been trained on the training set. According to empirical research, using 20%–30% of the data for testing and the remaining 70%–80% for training yields the best results. The resulting accuracy estimates for this division are valid—that is, they do not overestimate the accuracy (i.e., do not underestimate the approximation error)—and more accurate—that is, their approximation error overestimation is as low as it can be.

In many real situations, there is a model for a physical phenomenon, a model with several unknowable parameters. The model needs to be trained to determine these parameters from the known observations. It could seem that using all of the available data points to make this determination is the best way to estimate the model’s parameters. This is a wise decision if it is known for certain that the model accurately captures the relevant phenomenon. However, in reality, it is frequently questionable if the current model is actually suitable. If all the available data are simply used in these circumstances to establish the parameters of the model, overfitting frequently occurs, where the model accurately explains all the data without really being sufficient. It is advised that the observations be divided into training and testing data to prevent overfitting; to start, the parameters of the model are determined using the training data. Following that, the model’s predictions for each testing data point are compared with what was actually observed, and this comparison is used to determine the model’s accuracy ( 48 ).

Sample Design

The required sample size was determined using guidelines from Roscoe ( 49 ), according to which most behavioral research should employ sample sizes larger than 30 and less than 500 to avoid Type II errors ( 50 ). Type II errors occur when the null hypothesis is falsely rejected, and the hypothesis test fails to find any significant differences despite their existence. As a result, the null hypothesis is not rejected ( 51 ).

In this study, Facebook and WhatsApp were used to distribute the survey, which was also circulated via Google forms. The survey was carried out between the months of October and November 2020, when there was a partial shutdown strategy in accordance with regulations during the COVID-19 pandemic, and it was open until no further responses were received.

Traveler satisfaction can be measured using either traveler expectations or traveler perceptions. The former is typically investigated for proposed public transportation systems, whereas studies on the latter offer information for evaluating and improving existing systems ( 52 ). As in many developing nations, travelers in Amman are unfamiliar with BRT systems. To identify the effects of present public transportation system performance as perceived by individuals on an unrealized planned BRT system, travelers’ satisfaction with currently operating public transportation services was investigated. A well-designed BRT system is anticipated to keep current public transportation users and may even promote a transition from automobiles to public transportation ( 53 ). In Amman, there is a lack of knowledge about how BRT characteristics and rider socioeconomics affect rider satisfaction levels, particularly from the perspective of a bus rider. By answering this question, the study aims to develop a thorough understanding of how people perceive the BRT service.

The choice of variables utilized in this survey was based on earlier research with a comparable scope ( 24 , 28 , 54 ). Questions were asked to respondents on their socioeconomic characteristics as well as characteristics of their travel habits. Based on the current state of public transportation services, the primary goal of this survey was to identify the factors that could contribute to the willingness to use the proposed BRT system.

Residents of Amman city were the target group, therefore any responses from residents of other cities were disregarded. As a result, 19 responses were disregarded. As a consequence, a total of 238 replies were collected, and calculations using Equation 1 show that the margin of error (MOE) is 6.352% ( 55 , 56 ).

The MOE is defined as in Equation 13:

MOE = Z \times \sqrt{\frac{\hat{p} \times (1 - \hat{p})}{n}} \times 100

(13)

where Z is z score for a confidence level of 95% which is 1.96 as obtained from statistical tables; $\hat{p}$ is the sample proportion which is assumed to be 0.50 since this would be the maximum sample size; n is the sample size, which in this case is 238.

The MOE indicates that in 95 out of every 100 samples, the population proportion will fall between the range of 6.352% below and above the sample proportion of 0.5. This value represents the precision level and the range in which an estimate’s true value (such as the population mean or the proportion of respondents) may be located. The reliability of the aggregated data increases with the reduction in the MOE.

Results and Discussion

To learn more about the travel preferences and habits of a sample of Amman residents, a stated preference survey was designed and distributed to all participants.

In transportation planning two main types of surveys are usually used: revealed preference surveys which are used to capture the actual behavior of the traveler and stated preference surveys which are usually used when it is either too expensive or impractical to implement a revealed preference, which looks at actual behavior ( 57 ). Stated preference surveys are used to predict the behavior of the traveler based on a hypothetical situation ( 58 ). The assumption that people may not always keep on with their intentions is a common concern about the use of stated preference data. As a result, validity is a major issue with stated preference data.

Despite this, while it is true that stated preference surveys have their own drawbacks, they are still a viable option with great advantages in encountering the difficulties of estimating the behavior of travelers for new alternative transportation systems, especially if the new alternative is different than existing ones ( 60 ). Conversely, revealed preference studies have their own limitations such as focusing on real-world behavior and that the factors they examine may be highly collinear. Additionally, revealed preference studies are unable to identify accurately the alternatives that were not selected, which is another area where they frequently fall short ( 60 ).

Pearmain et al. ( 59 ) reviewed several studies examining the validity of predictions of choice behavior based on stated preference surveys. They found that the majority of these research findings were positive, showing that stated preference methods may accurately predict choice behavior for the sample under study. To this end, stated preference is still a viable option with many advantages for overcoming the challenges of predicting traveler behavior for new alternative transportation systems, especially if the new alternative differs from existing ones ( 60 ). For the purposes of this study, “stated preference survey” will be referred to as “survey” for notational convenience.

The survey that the participants filled out included questions on their socioeconomic characteristics, travel habits, and willingness to use BRT. The survey’s first part makes an effort to better understand the characteristics of the respondents, including gender, marital status, age, level of education, job, household monthly income, and the number of household automobiles. The second part of the survey aims to learn more about the features of people’s most significant daily trips, including the average number of trips per day, the purpose of the trip, the number of transfers made, the mode of transportation utilized, and the total travel time. The final part of the survey was used to learn about the travel habits of captive users, also referred to as transit dependent ( 61 ) or disadvantaged users ( 62 ) (i.e., lower-income individuals compelled to use public transportation because they do not own a car), as well as to learn about the travel habits of choice users, also referred to as discretionary users ( 63 ) (i.e., individuals who do own cars). The travel habits of these two user types were collected in this final part and are as follows: whether the individual uses the public transportation system or not, weekly frequency of public transportation system usage, willingness to use BRT, and for what purpose the individual prefers to use the BRT.

Survey Results and Analysis

The survey results as reported by respondents are shown in Table 1. The survey findings were initially checked for questionable information.

Table 1.

Variables Included in the Survey With Counts and Percentages Obtained Under Each Variable

Variable (abbreviation)	Categories	Count	Percentage
Part 1—Individuals’ characteristics
Gender	Female	132	55.5
Gender	Male	106	44.5
Marital status (status)	Single	126	52.9
	Married	108	45.4
	Other	4	1.7
Age	[18–24)	54	22.7
	[24–40)	129	54.2
	[40–50)	22	9.2
	[50–60)	17	7.1
	≥60	16	6.7
Education	High school diploma	16	6.7
	University student	10	4.2
	Associate degree	23	9.7
	BSc	112	47.1
	MSc	63	26.5
	PhD	9	3.8
	Other	5	2.1
Job	Two-year college student	2	0.8
	University student	27	11.3
	Private sector employee	63	26.5
	Public sector employee	26	10.9
	Self-employed	66	27.7
	Out of work	54	22.7
Salary (JD per month)	<400	85	35.7
	[400–600)	65	27.3
	[600–1,000)	49	20.6
	[1,000–2,000)	30	12.6
	>2,000	9	3.8
No. of automobiles owned (autos)	0	34	14.3
	1	84	35.3
	2	62	26.1
	3	46	19.3
	>3	12	5.0
Part 2—Individuals’ travel behavior characteristics
Average no. of daily trips (trips)	0	13	5.5
	1	56	23.5
	2	105	44.1
	3	29	12.2
	>3	35	14.7
Trip purpose(purpose)	Home-based-education (HBE)	29	12.2
	Home-based-work (HBW)	96	40.3
	Home-based-other (HBO)	86	36.1
	Non-home-based (NHB)	4	1.7
	Other	23	9.7
No. of transfers (transfer)	1	185	77.7
	2	38	16.0
	3	7	2.9
	>3	8	3.4
Travel time from end-to-end (time, minutes)	≤ 10	18	7.6
	(10–20]	76	31.9
	(20–30]	55	23.1
	(30–40]	28	11.8
	(40–50]	18	7.6
	(50–60]	15	6.3
	(60–90]	14	5.9
	(90–120]	3	1.3
	> 120	11	4.6
Part 3—Characteristics of captive and choice riders
Public transportation usage (usage)	Yes	72	30.3
Public transportation usage (usage)	No	166	69.7
Frequency of public transportation usage/week (frequency)	0	166	69.7
	1	4	1.7
	2	9	3.8
	3	8	3.4
	4	2	0.8
	5	12	5.0
	>5	14	5.9
	Missing values	23	9.7
Mode of transport used (used_mode)	Bus	27	11.3
	Taxi	39	16.4
	White cab	6	2.5
	Do not use public transportation	166	69.7
Knowledge of BRT (BRT_know)	Yes	236	99.2
Knowledge of BRT (BRT_know)	No	2	0.8
Willing to use BRT (will_BRT)	Yes	118	49.6
Willing to use BRT (will_BRT)	No	120	50.4
Purpose of proposed usage of BRT (BRT_purpose)	University	23	9.7
	Work	47	19.7
	Shopping	23	9.7
	Medical	4	1.7
	Social visits	13	5.5
	Entertainment	28	11.8
	Other	100	42.0
Approximate distance to BRT line (distance, km)	(0–1]	24	10.1
	(1–2]	19	8.0
	(2–5]	23	9.7
	>5	78	32.8
	I do not know where the BRT line is located	94	39.5

Note: BRT = bus rapid transit.

After the variable “Frequency” was thoroughly filtered out, 23 records were missing. Missing values indicate that the respondents did not provide any responses. It was decided not to delete the responses in these 23 records under the variable “Frequency” to prevent the loss of these records and to impute their values instead when developing the model.

According to Table 1 part 1, females represented 55.5% of the whole sample, and 52.9% of the individuals were single. Most of the sample were in the age range 24 to 40 years (54.2%), and 47.1% of the sample held a bachelor’s degree. About 65% are workers and 35.7% belonged to households with monthly salaries of 400 JD (USD 564) or less and about 35.3% belonged to households with one automobile.

Table 1 part 2 displays the findings for the respondents’ travel behavior characteristics. It reveals that 44.1% of the sample made two daily trips, and 40.3% of trips were home-based-work (HBW). In their most significant daily trip, 77.7% of individuals use just one transfer, and for their longest daily trip, 63% of individuals needed less than or equal to half an hour to complete it all from end to end.

The third part of Table 1 displays the findings for characteristics of captive and choice riders. It was revealed that 30.3% of those surveyed used public transportation for their most significant daily trips, with 68% of them utilizing it five times or less weekly. For the most significant daily trip, it was revealed that 16.4% of the sample utilized taxis, and that of those who used the public transportation system, around 54% used taxis over buses and white cabs. The majority of respondents were aware that there was a BRT project, and they were quite evenly divided between willingness and unwillingness to use BRT. While just 29% of responses were related to work or university trip, approximately 71% of respondents reported that they might use BRT for other trip purposes.

When respondents were asked about how far their homes or places of work were from the BRT line, it was revealed that 59% of them could estimate the distance. Only 10% of the respondents indicated that the distance to the BRT line was less than or equal to 1 km, while 32.8% estimated the distance to be greater than 5 km.

Cross-tabulation was used to examine the respondents’ responses about their willingness to use BRT for daily trips as well as the relevant factors that influence that willingness, as shown in Table 2.

Table 2.

Cross-Tabulations of Survey Responses Based on Willingness to Use Bus Rapid Transit (BRT)

	Willingness to use BRT
	Classes
	No		Yes
Categories	Count	% within category^*	Count	% within category
Gender
Female	50	38%	82	62%
Male	70	66%	36	34%
Marital status
Single	67	53%	59	47%
Married	50	46%	58	54%
Other	3	75%	1	25%
Age
[18–24)	10	19%	44	81%
[24–40)	82	64%	47	36%
[40–50)	13	59%	9	41%
[50–60)	7	41%	10	59%
≥60	8	50%	8	50%
Education
High school diploma	4	25%	12	75%
Student	3	30%	7	70%
Associate degree	7	30%	16	70%
BSc	56	50%	56	50%
MSc	43	68%	20	32%
PhD	6	67%	3	33%
Other	1	20%	4	80%
Job
Two-year college student	1	50%	1	50%
University student	5	19%	22	81%
Private sector employee	29	46%	34	54%
Public sector employee	13	50%	13	50%
Self-employed	54	82%	12	18%
Out of work	18	33%	36	67%
Salary (JD per month)
<400	25	29%	60	71%
[400–600)	43	66%	22	34%
[600–1,000)	29	59%	20	41%
[1,000–2,000)	21	70%	9	30%
≥2,000	2	22%	7	78%
No. of automobiles owned
0	8	24%	26	76%
1	33	39%	51	61%
2	32	52%	30	48%
3	39	85%	7	15%
>3	8	67%	4	33%
Average no. of daily trips
0	8	62%	5	38%
1	17	30%	39	70%
2	60	57%	45	43%
3	15	52%	14	48%
>3	20	57%	15	43%
Trip purpose
HBE	7	24%	22	76%
HBW	49	51%	47	49%
HBO	49	57%	37	43%
NHB	3	75%	1	25%
Other	12	52%	11	48%
No. of transfers
1	114	62%	71	38%
2	4	11%	34	89%
3	2	29%	5	71%
>3	0	0%	8	100%
Travel time from end-to-end (minutes)
≤10	8	44%	10	56%
(10–20]	54	71%	22	29%
(20–30]	24	44%	31	56%
(30–40]	13	46%	15	54%
(40–50]	12	67%	6	33%
(50–60]	4	27%	11	73%
(60–90]	3	21%	11	79%
(90–120]	0	0%	3	100%
>120	2	18%	9	82%
Public transportation usage
No	103	62%	63	38%
Yes	17	24%	55	76%
Frequency of public transportation usage/week
0	103	62%	63	38%
1	1	25%	3	75%
2	3	33%	6	67%
3	1	13%	7	88%
4	1	50%	1	50%
5	2	17%	10	83%
>5	1	7%	13	93%
Missing	23
Modes of transport used
Bus	3	11%	24	89%
Taxi	13	33%	26	67%
White cab	1	17%	5	83%
Do not use	103	62%	63	38%
Knowledge of BRT
No	1	50%	1	50%
Yes	119	50%	117	50%
Purpose of proposed usage of BRT
University	2	9%	21	91%
Work	11	23%	36	77%
Shopping	7	30%	16	70%
Medical	0	0%	4	100%
Visiting friends	5	38%	8	62%
Entertainment	10	36%	18	64%
Other	85	85%	15	15%
Approximate distance to BRT line (km)
≤1	10	42%	14	58%
(1–2]	8	42%	11	58%
[2–5]	12	52%	11	48%
>5 km	56	72%	22	28%
I do not know where the BRT line is located	34	36%	60	64%
Total in sample	120		118

Note: HBE = home-based-education; HBW = home-based-work; HBO = home-based-other; NHB= non-home-based.

$% within variabl e^{'} s specific category = \frac{count ofspecific category who belong to a specific class}{total specific category sample} * 100 %$

As shown in Table 2, 62% of the females in the sample expressed a willingness to use BRT, compared with 34% of the males. With regard to marital status, 47% of single people indicated a willingness to use BRT, and 54% of those who were married.

Eighty-one percent of respondents between the ages of 18 and 24 said they would be willing to use BRT, compared with 59% of those aged 50 to 59. Of those in the age ranges of 60 and above, 40 to 49, and 24 to 39, respectively, 50%, 41%, and 36% were willing to use BRT.

With regard to educational level, both willingness and unwillingness to use BRT were equal among BSc holders, however, lower proportions of those more highly educated (32% of MSc holders and 33% of PhD holders) were willing to use BRT. Other educational levels showed higher willingness to use BRT: 75% of high school graduates and 70% of students and associates.

With regard to occupation, employees in the public sector and two-year college students were evenly divided on being willing or unwilling to use BRT. Just 18% of business owners were willing to use BRT, compared with 81% of university students, 67% of unemployed individuals, and 54% of employees in the private sector.

Of respondents with monthly incomes under 400 JD (USD 564), 71% were willing to use BRT. Thirty-four percent of respondents with monthly income of 400–599 JD (USD 564–844) said they were willing to use BRT, compared with 41% of those with monthly income of 600–999 JD (USD 845–1,408), 30% of those with 1,000–1,999 JD (USD 1,409–2,816), and 78% of those with 2,000 JD or more (USD 2,817).

Of those without automobiles, 67% said they were willing to use BRT, compared with 61% with one car, 48% of two-car owners, 15% of those with three cars, and 33% of those who owned more than three cars.

Seventy percent of individuals who made one daily trip stated they were willing to use BRT, compared with 48% of those who commuted three times per day. Those who made two daily trips or more than three daily trips both indicated that they were willing to use BRT, with each group reporting 43%. The willingness to use BRT was 38% among individuals who did not make daily trips.

With regard to the trip purpose, 76% of respondents said they were willing to use BRT for home-based education (HBE), while 49% said they were willing to use BRT for home-based work (HBW); 43% and 25% said they were willing to use BRT for home-based-other (HBO) and non-home-based (NHB) purposes, respectively, while 48% of respondents said they would be using BRT for other purposes.

Eighty-nine percent of respondents who used two transfers were willing to use BRT, while 100% of those who used more than three transfers stated that they were willing to use BRT. Seventy-one percent of those who used three transfers and 38% of respondents who only used one transfer were willing to use BRT.

One hundred percent and 82% of respondents, respectively, indicated that they were willing to use BRT if it took them between 91 and 120 min and longer than 90 mins to perform their daily trip. Seventy-nine percent of individuals who needed 61 to 90 min to perform their trips were willing to use BRT. The willingness to use BRT was 73% among those who needed between 51 and 60 min. Fifty-six percent of individuals who traveled for less than 10 min as well as those who traveled between 21 and 30 min were willing to use BRT, whereas 54% of those who traveled between 31 and 40 min were willing to use BRT; 33% and 29%, respectively, of those who traveled between 41 and 50 min and between 11 and 20 min, were willing to use BRT.

Thirty-eight percent of those who did not use public transportation were willing to use BRT, compared with 76% of those who used public transportation.

Ninety-three percent of those who commuted by public transportation more than five times per week were willing to use BRT. Eight-eight percent of individuals who commuted by public transportation three times per week said they were willing to use BRT, compared with 83% of those who commuted by public transportation five times per week. The willingness to use BRT was 67% among individuals who commuted by public transportation twice each week, and 50% of individuals who commuted four times per week by public transportation said they were willing to use BRT.

Eighty-nine percent of of respondents who commuted daily by bus and 83% of those who used white cabs said they were willing to use BRT. Sixty-seven percent of those who used taxis indicated that they were willing to use BRT.

The survey’s findings also showed that 50% of respondents who knew about the BRT project were willing to using it. With regard to the purpose of proposed usage of BRT, 91% of respondents were willing to using BRT for university trips, 77% for work trips, 70% for shopping trips, 64% for entertainment trips, 62% for visiting friends, and 15% for other trip purposes.

The respondents were also asked to estimate the approximate distance from their place of employment or housing to the nearest BRT line. Sixty-four percent of those who did not know the approximate location of the nearest BRT line were willing to using it. If the distance were less than or equal to 1 km, 58% of respondents said they would use BRT. Moreover, 58% of the respondents in the survey who indicated that the distance was approximately larger than 1 to 2 km were willing to use BRT. Twenty-eight percent of those who estimated the distance to the nearest BRT line to be more than 5 km away were willing to use BRT.

According to the survey results section, there are 23 missing values for the variable “frequency of public transportation usage/week,” as can be seen in Table 1. It was decided to leave these records in place and impute their values using the unsupervised variable filter to fill in the missing data. When a record has missing data, the filter substitutes it with the mode of all known values of the variable in the class that the record belongs to. For example, there are 23 missing values under the variable “Frequency of public transportation usage/week,” if one of these missing cells’ values under this variable is located in row 20. If row 20 is classified to belong to “No” under the class variable “Willingness to use BRT,” the value of this missing cell data will be calculated based on the modes of other values in “Frequency of public transportation usage/week” that are classified as “No.” WEKA (3.9.4) is the software used to carry out the imputation ( 64 ).

Overall Model Evaluation

As shown in Table 1, the response variable (Willingness to use BRT) has two categories (yes and no), and both BLR and BNs were utilized to model the willingness to use BRT.

A BLR model was developed using stepwise regression, along with six BNs utilizing various combinations of search and score metrics. Each model was initially split into a 70% training sample and a 30% test sample. Training and testing were repeated for five runs, with the average of those runs being recorded. The outcomes of the models developed using BNs were then compared with the model developed using BLR based on evaluation measures. Performance measurements of accuracy, sensitivity, specificity, and area under the Reciever Operating Characteristic curve serve as the basis for the comparison.

The equations for the used performance measures are shown in Equations 14 to 16:

Accuracy = \frac{TO + T 1}{TO + FO + T 1 + F 1}

(14)

Sensitivity = \frac{TO}{TO + FO}

(15)

Specificity = \frac{T 1}{T 1 + F 1}

(16)

where TO denotes the number of individuals who are not willing to use BRT which are correctly classified; T1 denotes the number of individuals who are willing to use BRT which are correctly classified; FO denotes the number of individuals who are not willing to use BRT which are incorrectly classified, and F1 denotes the number of individuals who are willing to use BRT which are incorrectly classified. The area under the curve (AUC) is the area below a curve with sensitivity plotted on the y-axis and 1-specificity ( 65 ).

Table 3 shows the average results of the performance measures used for the different models developed, where a corrected paired t-test was used to test their statistical significance. To that end, 10 models were developed.

Table 3.

Comparison of Evaluation Measures in Binomial Logistic Regression (BLR) and Bayesian Network (BN) Models

		Performance measure				Complexity for BNs	No. of selected variables in model	No. of wins/losses
Modeling techniques		Accuracy	Sensitivity	Specificity	Area under curve	Complexity for BNs	No. of selected variables in model	No. of wins/losses
BLR model		0.751 (0.045)	0.778 (0.098)	0.723 (0.136)	0.858 (0.041)	-	12	1/0
BN models: Algorithm+Metric	Hillclimbing+Bdeu	0.742 (0.579)	0.694 (0.088)	0.791 (0.090)	0.798 (0.066)	16	15	1/2
	Hillclimbing+MDL	0.734 (0.0621)	0.700 (0.077)	0.768 (0.107)	0.807 (0.022)	16	15	0/0
	Hillclimbing+AIC	0.739 (0.064)	0.761 (0.117)	0.718 (0.037)	0.843 (0.047)	22	17	0/1
	K2+Bdeu	0.731 (0.0446)	0.717 (0.077)	0.747 (0.092)	0.849 (0.036)	14	14	0/0
	K2+MDL	0.740 (0.042)	0.744 (0.097)	0.735 (0.085)	0.854 (0.039)	13	13	1/0
	K2+AIC	0.762 (0.044)	0.789 (0.072)	0.734 (0.053)	0.877 (0.053)	24	17	2/1
	Simulated+Bdeu	0.720 (0.036)	0.744 (0.105)	0.695 (0.103)	0.832 (0.022)	17	15	0/1
	Simulated+MDL	0.714 (0.047)	0.706 (0.093)	0.723 (0.078)	0.811 (0.036)	15	15	0/1
	Simulated+AIC	0.767 (0.054)	0.756 (0.087)	0.780 (0.073)	0.863 (0.027)	25	17	1/2

Note: MDL = minimum description length; AIC = Akaike information criterion.

Numbers in parenthesis are standard deviations.

Complexity for BNs refers to the number of arcs.

Bold values indicate the best performance obtained within the measure.

Underlined values indicate the worst performance obtained within the measure.

The accuracy for the BLR model was in the range of the accuracies attained by BNs, according to results derived from the developed models. The BLR model had the second-highest accuracy (0.751), while simulated+AIC had the highest accuracy (0.767) and simulated+MDL had the lowest accuracy (0.714).

As shown in Table 3, there was no statistically significant difference in the performance of the BLR model and all of the BNs used when applying the corrected paired t-test. The values for sensitivity ranged from 0.694 for hillclimber+Bdeu to 0.789 for K2+AIC. The results also showed that there was no statistically significant difference in the BLR model’s performance compared with all other BNs when utilizing the corrected paired t-test.

For specificity, K2+AIC obtained the highest value of 0.791, while simulated+Bdeu obtained the lowest value of 0.695. According to the results of the paired t-test, there was no statistically significant difference between the BLR model and the employed BNs. For AUC, the findings showed that K2+AIC had the highest value of 0.863 and hillclimbing+Bdeu had the lowest value of 0.798. However, when paired t-test was employed to determine statistical significance, all of the models were comparable with the BLR model.

The models were compared using two additional metrics: the number of selected variables and the complexity of the BN model, which is the number of arcs in the developed BN tree, because the values obtained by the evaluation measures did not have significant levels of variability. Table 3 displays the findings of this comparison with the number of wins and losses in each column. As can be seen, the BLR model, which had the least number of selected variables in the constructed model and a single win, was disqualified from further analysis. The only model that will be analyzed is K2+AIC, which had two wins, indicating that BNs outperformed the BLR model.

Discussion

The developed BN using K2+AIC algorithm was used for further analysis and to determine which variables contribute to the willingness to use BRT. Age, gender, the number of transfers, and the purpose of the proposed use of BRT were shown to be directly related to willingness to use BRT when utilizing K2+AIC to develop BN, as shown in Figure 2.

Figure 2.

Bayesian network (BN) model tree using K2+AIC.

The inference method, in which evidence was set to a specific category of the variables that were tied directly by an arc with willingness to use BRT, was used to identify the most significant variables that were associated with willingness to use BRT. The probability of willingness to use BRT was calculated as shown in Table 4.

Table 4.

Inference Results for Categories of Variables in the Selected BNs and Their Interdependence With BRT-Will

Categories of variables with evidence	Probability of using BRT in K2+AIC
BRT-Purpose
University	0.8752
Work	0.7415
Shopping	0.6720
Medical	0.8252
Social visits	0.5708
Entertainment	0.6422
Transfer
2	0.8692
3	0.6867
4	0.8782
Autos
0	0.5556*
1	0.5089*
Age
[18-24)	0.8123
[50-60)	0.5678
Gender
Female	0.6202

Note: BRT = bus rapid transit; BN = Bayesian network; AIC = Akaike information criterion. Bold numbers indicate that the values obtained are the highest for the respective variable and/or category of variable when compared with others. *Variable is not directly connected to BRT-will.

Using the conditional probability table of the BN developed, these probabilities were calculated, for instance, by giving a probability of one to the value age [18–24) = 1, which leads to a probability of will-BRT of 0.8752 in the K2+AIC model.

It was found that the probability of using BRT increased with the number of transfers. This suggests that people anticipate BRT to shorten their daily trip distance by reducing the number of transfers along their itinerary. This outcome was in accordance with Redman et al. ( 66 ) who found that if trips were reduced by 15 min, rail services would increase by 25%. Additionally, Eriksson et al. ( 67 ) revealed, using a post-trip questionnaire survey of automobile users, that one of the most important factors to increase the attractiveness of public transportation is shorter travel time.

It was found that households without cars were more likely to be willing to use BRT. This was found to be consistent with the findings of Paulley et al. ( 68 ), who showed that as income increases, it is expected that both the number of trips and their average length will increase. They expected that if automobile availability was low and public transportation was good, the increased number of trips would be made using public transportation, but if car availability was high, the trips would be made using private vehicles.

In K2+AIC, “university” was identified as the primary BRT-Purpose category that was associated with a higher likelihood of will-BRT. Using BRT for university or educational trips was consistent with the conclusion of Eriksson and Forward ( 69 ) that shopping, leisure, and work or study trips are the most frequent trip purposes. Additionally, Zailani et al. ( 70 ) found in their study of residents in Malaysia’s capital that trips for education or work by public transportation were more frequent than those for shopping and leisure.

The probability of using BRT was found to be highest in the age group between 18 and 23, which also explains why university trips were one of the most important anticipated trip purposes for BRT. The outcome was similar in Kim and Wang’s ( 71 ) study of the factors affecting travel behavior at the individual and neighborhood levels, which demonstrated using multilevel regression that the probability of using a car increased as people’s ages and incomes increased.

The largest significant increase in the probability of BRT usage was seen among females, which was an intriguing, surprising finding. This finding contrasts with that of Aloul et al. ( 72 ), who found that 47% of Jordanian women surveyed rejected employment offers because of the poor quality of public transportation, including its availability and affordability, and the risk of sexual harassment. According to Aloul et al. ( 72 ), female commuters believe that their economic status is strongly associated with their level of mobility, and more than 80% of the women surveyed think that public transportation is essential for women’s economic empowerment. A recent report that was issued by The World Bank ( 7 ) pointed out that one of the difficulties that faces the public transportation system in Jordan is being unsafe especially for females. Only 50% of females in Jordan use public transportation, making up only one-third of all users.

However, this study’s conclusions on the potential increase in females’ utilization of the proposed BRT are consistent with those of Fuiji and Kitamura ( 73 ). They claimed that if a shift in the temporal structure positively affected attitudes about utilizing public transportation, this may eventually result in use of public transportation on a permanent basis. The findings here and those of Fuiji and Kitamura ( 73 ) are similar in that females anticipate BRT to provide higher-quality service than current public transportation options. It appears that Jordanian females support the country’s shift toward BRT.

Conclusion and Recommendations

This study examines how people perceive the BRT system using survey data and identifies the factors that affect people’s willingness to use the BRT. BLR and BNs were employed to analyze the survey results. It was found that the models developed with BNs were superior to those developed with BLR.

Inference was used to identify the most important variables influencing the willingness to use BRT. Evidence was set to categories of variables that were directly connected by an arc to the class variable (willingness to use BRT), and the probability of the class variable was then calculated.

The inference results indicated that respondents from households without automobiles were more likely to use BRT. Additionally, it was found that individuals anticipate BRT to reduce their daily trip distance by reducing the number of transfers they must make along their journey. University students were found to be the most willing to use BRT, while young travelers between the ages of 18 and 23 also showed the most willingness to use BRT. Interestingly, it was shown that females were most willing to use BRT. This shows that BRT is expected to function at a higher standard than present public transportation services, which might attract and motivate people to commute via public transportation. According to the findings, people anticipate that BRT will improve both their daily travel experience and the quality of the transportation services that are now offered.

The analysis conclusions can be used to provide several recommendations for increasing BRT ridership. Targeting Amman’s youth, especially university students, is essential given that travel to university is one of the most anticipated trips by BRT. In addition, more females are willing to utilize BRT because of its anticipated improved operational services, which would result in an increase in female riders.

Third, it is suggested that BRT services be subsidized for households without privately owned vehicles or, more specifically, for low-income households, because these households are more likely to use BRT. Furthermore, the survey’s findings showed that greater publicity and marketing are required because 40% of respondents were unaware of where the BRT line was located.

The findings presented in this paper are crucial for planners and decision-makers in public transportation, especially in communities with characteristics comparable to Amman. Any society with a culture of public transportation that is negatively affected by a variety of variables is included in the intended meaning of the Amman profile. These variables can include irregular timing and routing, poorly marked bus stops, daily fluctuations in the number of buses for each route, a lack of transportation planning studies, and auto-oriented neighborhoods in areas with comparable economic condition to Jordan.

It was evident that, even in communities where cars are the norm, adding new systems to the existing public transportation services is well received. To promote the use of public transportation services and discourage the use of automobiles, it is essential to monitor and evaluate the mobility demands of people.

Footnotes

Author Contributions

The authors confirm contribution to the paper as follows: study conception and design: R. Mujalli, R. Alawadi; data collection: M. Al-Kasasbeh, A. Tarawneh; analysis and interpretation of results: R. Mujalli; draft manuscript preparation: R. Mujalli. All authors reviewed the results and approved the final version of the manuscript.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iDs

Randa Oqab Mujalli

Maha Al-Kasasbeh

Data Accessibility Statement

Data that support the findings of this study are available from the corresponding author on reasonable request.

References

Deng

Nelson

J. D.

The Perception of Bus Rapid Transit: A Passenger Survey from Beijing Southern Axis BRT Line 1. Transportation Planning and Technology, Vol. 35, No. 2, 2012, pp. 201–219.

Deng

Nelson

Recent Developments in Bus Rapid Transit: A Review of the Literature. Transport Reviews, Vol. 31, No. 1, 2011, pp. 69–96.

Shalan

Analyzing the Public Transportation in Amman: The Case of the Bus Rapid Transit (BRT), Ph.D. Thesis. University of Stuttgart, Stuttgart, 2013.

The World Bank. Jordan Country Reclassification - Questions and Answers, The World Bank Group, Washington, D.C., 2017.

Land Transport Regulatory Commission. Quarter Annual Report for the Months October, November, and December, 2020. https://ltrc.gov.jo/sites/default/files/ltqryr_lrby_lrb_2020_.pdf. Accessed January 30, 2022.

Department of Statistics. Estimated Population of the Kingdom by Governorate and Sex, at End-year 2021, 2021. https://jorinfo.dos.gov.jo/Databank/pxweb/en/Demographi_Statistics/Demographi_Statistics/Table1.px/. Accessed October 9, 2022.

The World Bank. Jordan Public Transport Diagnostic Recommendations. The World Bank Group, Washington, D.C., 2022.

Gwilliam

Urban Transport in Developing Countries. Transport Reviews, Vol. 23, 2003, pp. 197–216.

Land Transport Regulatory Commission. Quarter Annual Report for the Months October, November, and December, 2021. https://ltrc.gov.jo/sites/default/files/mwshrt_ld_lsn_2021_lrb_lrb_rkb__0.pdf. Accessed October 20, 2022.

10.

Department of Statistics. Average of Annual Current Household Income by Source of Income and Governorate and Urban/Rural (JOD), 2017. http://www.dos.gov.jo/dos_home_a/main/linked-html/household/2017/G3/Table3G3_King.pdf. Accessed February 7, 2022.

11.

Greater Amman Municipality. The Bus Rapid Transit (BRT) Project, 2022. http://www.ammanbrt.jo/?l=en#:∼:text=The%20concept%20revolves%20around%20large-capacity%20buses%20that%20operate,of%20a%20fully-integrated%20and%20hierarchical%20public%20transport%20network. Accessed October 18, 2022.

12.

Al-Sharari

Assessing the Public Participation Initiatives in the Planning and Delivery of the Amman Bus Rapid Transit Project. Case Studies on Transport Policy, Vol. 10, No. 2, 2022, pp. 1314–1324.

13.

Al-Dalain

Alnsour

Measuring the Sustainability Performance of the Public Transport System: A Case of the Amman BRT System in Operation. International Journal of Project Management and Productivity Assessment, Vol. 10, 2022, pp. 1–13.

14.

Stojic

Ciric

Sedlak

Horvat

Students’ Views on Public Transport: Satisfaction and Emission. Sustainability, Vol. 12, No. 20, 2020, pp. 1–16.

15.

Eboli

Mazzulla

Willingness-to-Pay of Public Transport Users for Improvement in Service Quality. European Transport, Vol. 38, 2008, pp. 107–118.

16.

Eboli

Mazzulla

How to Capture the Passengers’ Point of View on Transit Service Through Rating and Choice Options. Transport Reviews, Vol. 30, No. 4, 2010, , pp. 435–450.

17.

Hensher

Stopher

Bullock

Service Quality: Developing a Service Quality Index in the Provision of Commercial Bus Contracts. Transportation Research: Part A, Vol. 37, 2003, pp. 499–517.

18.

Dell’Olio

Ibeas

Cecın

Modelling User Perception of Bus Transit Quality. Transport Policy, Vol. 17, No. 6, 2010, pp. 388–397.

19.

De Oña

López

Transit Service Quality Analysis Using Cluster Analysis and Decision Trees: A Step Forward to Personalized Marketing in Public Transportation. Transportation, Vol. 43, 2016, pp. 725–747.

20.

Nguyen-Phuoc

D. Q.

Tran

A. T. P.

Nguyen

T. V.

P. T.

D. N.

Investigating the Complexity of Perceived Service Quality and Perceived Safety and Security in Building Loyalty Among Bus Passengers in Vietnam – A PLS-SEM Approach. Transport Policy, Vol. 101, 2021, pp. 162–173.

21.

De Oña

Eboli

Mazzulla

Perceived Service Quallity in Bus Transit Service: A Structural Equation Approach. Transport Policy, Vol. 29, 2013, pp. 219–226.

22.

Stein

Morris

Hall

Nock

Structural Equation Modelling. Methods and Protocols, Vol. 1666, 2017, pp. 557–580.

23.

De Oña

Calvo

A Classification Tree Approach to Identify Key Factors of Transit Service Quality. Expert Systems with Applications, Vol. 39, 2012, pp. 11164–11171.

24.

De Oña

Analyzing Transit Service Quality Evolution Using Decision Trees and Gender Segmentation. WIT Transactions on the Built Environment, Vol. 130, 2013, pp. 611–621.

25.

De Oña

Eboli

Mazzulla

Heterogeneity in Perceptions of Service Quality Among Groups of Railway Passengers. International Journal of Sustainable Transportation, Vol. 9, No. 8, 2014, pp. 612–626.

26.

De Oña

Eboli

Mazzulla

Key Factors Affecting Rail Service Quality. A Decision Tree Approach. Transport, Vol. 29, No. 1, 2014, pp. 75–83.

27.

Garrido

De Oña

Neural Networks for Analysing Service Quality in Public Transportation. Expert Systems with Applications, Vol. 4, 2014, pp. 6830–6838.

28.

De Oña

Garrido

Extraction of Attribute Importance from Satisfaction Surveys with Data Mining Techniques: A Comparison between Neural Networks and Decision Trees. The International Journal of Transportation Research, Vol. 9, No. 1, 2016, pp. 39–48.

29.

Mujalli

De Oña

A Method for Simplifying the Analysis of Traffic Accidents Injury Severity on Two-Lane Highways Using Bayeisan Networks. Journal of Safety Research, Vol. 2, 2011, pp. 317–326.

30.

De Oña

Mujalli

Calvo

Analysis of Traffic Accident Injury Severity on Spanish Rural Highways Using Bayesian Networks. Accident Analysis and Prevention, Vol. 43, 2011, pp. 402–411.

31.

De Oña

Lopez

Mujalli

Calvo

Analysis of Traffic Accidents on Rural Highways Using Latent Class Clustering and Bayesian Networks. Accident Analysis and Prevention, Vol. 51, 2013, pp. 1–10.

32.

Mujalli

Lopez

Garach

Bayes Classifiers for Imbalanced Traffic Accidents Datasets. Accident Analysis and Prevention, Vol. 88, 2016, pp. 37–51.

33.

Mujalli

Modeling Risk of Road Crashes Using Aggregated Data. Jordan Journal of Civil Engineering, Vol. 12, No. 1, 2018, pp. 45–60.

34.

Mujalli

Garach

Lopez

Al-Rousan

Evaluation of Injury Severity for Pedestrian-Vehicle Crashes in Jordan Using Extracted Rules. Journal of Transportation Engineering, Part A: Systems, Vol. 145, No. 7, 2019, pp. 1–13.

35.

Perucca

Salini

Travellers’ Satisfaction with Railway Transport: A Bayesian Network Approach. Quality Technology & Quantitative Management, Vol. 11, No. 1, 2014, pp. 71–84.

36.

Yang

Cheng

Hou

Evaluating the Service Quality of Public Transit from Passengers’ Perceptions Using Bayesian Networks. Presented at 93rd Annual Meeting of the the Transportation Research Board, 14-1900, 2014.

37.

Yang

Rasouli

Exploring Passenger Assessments of Bus Service Quality Using Bayesian Networks. Journal of Public Transportation, Vol. 19, No. 3, 2016, pp. 36–54.

38.

Mandhani

Nayak

J. K.

Parida

Interrelationships Among Service Quality Factors of Metro Rail Transit System: An Integrated Bayesian Networks and PLS-SEM Approach. Transportation Research Part A Policy and Practice, Vol. 140, 2020, pp. 320–336.

39.

Díez-Mesa

De Oña

Bayesian Networks and Structural Equation Modelling to Develop Service Quality Models: Metro of Seville Case Study. Transportation Research Part A: Policy and Practice, Vol. 118, 2018, pp. 1–13.

40.

Neapolitan

R. E.

Learning Bayesian Networks. Prentice Hall, Upper Saddle River, NJ, 2004.

41.

Cooper

G. F.

Herskovits

A Bayesian Method for the Induction of Probabilistic Networks from Data. Machine Learning, Vol. 9, 1992, pp. 309–347.

42.

Madden

On the Classification Performance of TAN and General Bayesian Networks. Knowledge-Based Systems, Vol. 22, 2009, pp. 489–495.

43.

Russell

S. J.

Norvig

Artificial Intelligence: A Modern Approach. Prentice Hall, Upper Saddle River, NJ, 2004.

44.

Arora

J. S.

Introduction to Optimum Design. In Discrete Variable Optimum Design Concepts and Methods: Simulated Annealing. Elsevier, Iowa City, Iowa, 2012, pp. 619–641.

45.

Bouckaert

R. R.

Bayesian Belief Networks: From Construction to Inference. Ph.D. Thesis. University of Utrecht, The Netherlands, 1995.

46.

Müller-Putz

Scherer

Brunner

Leeb

Pfurtscheller

Better Than Random? A Closer Look on BCI Results. International Journal of Bioelectromagnetism, Vol. 10, No. 1, 2008, pp. 52–55.

47.

Ben-Akiva

Lerman

Discrete Choice Analysis: Theory and Application to Travel Demand. MIT Press, Cambridge, MA, 1985.

48.

Gholamy

Kreinovich

Kosheleva

Why 70/30 or 80/20 Relation Between Training and Testing Sets: A Pedagogical Explanation, 2018. Scholar Works @UTEP. https://www.scribbr.com/apa-examples/report/. Accessed October 1, 2022.

49.

Roscoe

J. T.

Fundamental Research Statistics for the Behavioral Sciences, 2nd ed. Holt Rinehart and Winston, New York, 1975.

50.

Sekaran

Bougie

Research Methods for Business: A Skill Building Approach, 5th ed. John Wiley & Sons Ltd, West Sussex, UK, 2010.

51.

Montgomery

D. C.

Runger

G. C.

Applied Statistics and Probability for Engineers, 5th ed. John Wiley & Sons, Inc., Danvers, MA, 2011.

52.

Wan

Kamga

Liu

Sugiura

Beaton

E. B.

Rider Perception of a “Light” Bus Rapid Transit System - The New York City Select Bus Service. Transport Policy, Vol. 49, 2016, pp. 41–55.

53.

Litman

A New Transit Safety Narrative. Journal of Public Transportation, Vol. 17, No. 4, 2014, pp. 114–135.

54.

Al-Shaar

Nehme

Bonin

Adjizian-Gerard

Al-Shaar

Passengers Receptivity of a New Public Transport Mode: Case of a BRT Project in Lebanon. Computational Urban Science, Vol. 2, No. 25, 2022, pp. 1–13.

55.

Cochran

W. G.

Sampling Techniques, 3rd ed. John Wiley & Sons, New York, 1977.

56.

Moore

D. S.

McCabe

G. P.

Introduction to the Practice of Statistics. W. H. Freeman, New York, 1999.

57.

Haider

Stated Preference and Choice Models–A Versatile Alternative to Traditional Recreation Research. Bodenkultur University, Vienna, Austria, 2002.

58.

Lee-Gosselin

The Scope and Potential of Interactive Stated-Response Data Collection Methods. Transportation Research Board, Irvine, CA, 1995.

59.

Pearmain

Swanson

Kroes

Bradley

Stated-Preference Techniques: A Guide to Practice. Steer Davies Gleave and Hague Consulting Group, London, 1991.

60.

Federal Highway Administration. Stated-Preference Surveys. U.S. Department of Transportation, Washington, D.C., 1996.

61.

Polzin

Chu

Rey

Density and Captivity in Public Transit Success: Observations from the 1995 Nationwide Personal Transportation Study. Transportation Research Record: Journal of the Transportation Research Board, 2000. 1735(1): 10–18.

62.

Litman

You Can Get There From Here: Evaluating Transportation Choice. Transportation Research Record: Journal of the Transportation Research Board, 2001. 1756: 32–41.

63.

Giuliano

Low Income, Public Transit, and Mobility. Transportation Research Record: Journal of Transportation Board, 2005. 1927(1): 63–70.

64.

Frank

Hall

Witten

The WEKA Workbench. Online Appendix for Data Mining: Practical Machine Learning Tools and Techniques, 4th edn. Morgan Kaufmann, Burlington, MA, 2016.

65.

Bradley

A. P.

The Use of the Area Under the ROC Curve in the Evaluation of Machine Learning Algorithms. Pattern Recognition, Vol. 30, No. 7, 1997, pp. 1145–1159.

66.

Redman

Friman

Gärling

Hartig

Quality Attributes of Public Transport that Attract Car Users: A Research Review. Transport Policy, Vol. 25, 2013, pp. 119–127.

67.

Eriksson

Friman

Gärling

Stated Reasons for Reducing Work-Commute by Car. Transportation Research Part F: Traffic Psychology and Behaviour, Vol. 11, 2008, pp. 427–433.

68.

Paulley

Mackett

Preston

Wardman

Titheridge

White

The Demand for Public Transport: The Effects of Fares, Quality of Service, Income and Car Ownership. Transport Policy, Vol. 13, 2006, pp. 295–306.

69.

Eriksson

Forward

Is the Intention to Travel in a Pro-Environmental Manner and the Intention to Use Car Determined by Different Factors?

Transportation Research Part D: Transport and Environment, Vol. 16, 2011, pp. 372–376.

70.

Zailani

Iranmanesh

Masron

Chan

T.-H.

Is the Intention to Use Public Transport for Different Travel Purposes Determined by Different Factors?

Transportation Research Part D: Transport and Environment, Vol. 49, 2016, pp. 18–24.

71.

Kim

Wang

Empirical Examination of Neighborhood Context of Individual Travel Behaviors. Applied Geography, Vol. 60, 2015, pp. 230–239.

72.

Aloul

Naffa

Mansour

Gender in Public Transportation: A Perspective of Women Users of Public Transportation. Freidric Ebert Stiftung, Bonn, Germany, 2018.

73.

Fuiji

Kitamura

What Does a One-Month Free Bus Ticket Do to Habitual Drivers? An Experimental Analysis of Habit and Attitude Change. Transportation, Vol. 30, No. 1, 2003, pp. 81–95.