Analysis of Mode Choice Affects From The Introduction of Doha Metro Using Machine Learning and Statistical Analysis (2023)

Transportation Research Interdisciplinary Perspectives 20 (2023) 100852
Contents lists available at ScienceDirect
Transportation Research Interdisciplinary Perspectives

journal homepage: www.sciencedirect.com/journal/transportation-
research-interdisciplinary-perspectives
Analysis of mode choice affects from the introduction of Doha Metro using
machine learning and statistical analysis
Ammar Abulibdeh
Applied Geography and GIS Program, Department of Humanities, College of Arts and Sciences, Qatar University, Doha, Qatar
A R T I C L E I N F O A B S T R A C T
Keywords: The aim of this study was to investigate the possible influences of the operation of the new Doha Metro on the
Machine learning travel mode choice behavior in Doha City, Qatar. Revealed preference (RP) and stated preference (SP) survey
FIFA World Cup 2022 questionnaires were designed to collect the necessary data. The questions considered different trip conditions
Travel behavior
and socioeconomic factors of travelers. Three different mode choices were considered in this study: private cars,
Trip characteristic
Stated preference survey
taxi services, and metro. Two statistical models and one machine learning model were used to analyze the
Statistical analysis current and future mode choices: discrete choice binary logit (BL) and multinomial logit (MNL) models as well as
extreme gradient boosting (XGBoost). Furthermore, the SHapley Additive exPlanations (SHAP) method was used
to rank the input features based on their importance according to the mean SHAP value. The results showed that
the XGBoost model outperforms the other two models in terms of predicting the travel mode choice as well as in
terms of its accuracy. The results showed that various trip characteristics are significant in determining the mode
choice, including the number of travelers and bags, journey time, and reimbursement of parking fees. Further
more, different socioeconomic characteristics proved to be significant for the current and future mode choices,
including nationality, income, age, employment status, and vehicle ownership.
1. Introduction transportation, and as a solution to the problems of urban traffic

congestion and rapidly increasing travel demand. Metros offer unpar
Large urban centers often face significant transport problems, such as alleled quality of service (in terms of speed, frequency of service, and
traffic and road congestion, increased car ownership and automobile travel time reduction) and can act as a better mitigation approach to
dependence, emission-based pollution, and decreased quality of life. automobile travel than purely bus-based systems (Golias, 2002; Zaidan
Many cities across the globe have invested heavily in the introduction, & Abulibdeh, 2018). In many metropolitan areas, public transportation
extension, or redevelopment of old and new urban public transit systems systems have attracted more attention in recent years at the expense of
with the aim of managing/mitigating automobile dependence and its single-occupancy vehicles because of the restricted capacity of trans
associated problems. Many scholars view a rail-based system as a major portation infrastructure and the need to achieve sustainability (Zaidan
public transport option that can comprise a crucial element in incapa and Abulibdeh, 2021; Moeckel et al., 2014). These metropolitan areas
citating the substantial problems raised by the use of private automo have aimed to develop and enhance the public transport network so as to
biles (Chen et al., 2016; Halse et al., 2019; Ibrahim et al., 2020; Zhu increase ridership and at the same time, discourage individuals from
et al., 2020). Advocates of public transportation systems favor such using private vehicles for commuting. However, despite many efforts in
systems based on perceived benefits that mainly include increased this area, the share of single-occupancy vehicles is still much larger than
market share for public transportation, reduced environmental impact, that of public transport (Abulibdeh et al., 2015a; Abulibdeh et al.,
reduced automobile dependence, and positive effects on urban devel 2015b; Li et al. 2017; Ouda et al. 2013).
opment (Abulibdeh, 2017; Golias, 2002; Hawas et al., 2016). Travel demand forecasting for new transport systems, particularly
Globally, most efforts to develop new transportation systems have for new metro and subway systems before their implementation, is a key
been focused on metros (or subways). Implementing a rail-based transit undertaking in investigating travelers’ attitudes to – and likely behav
system, particularly metro rail, is seen by many cities in developing iors toward – new systems (Heinen et al., 2017; Liu et al., 2017; Sohoni
countries as a promising approach to achieving sustainable et al., 2017). Several studies have examined the impacts of new public
E-mail address: [email protected].
https://doi.org/10.1016/j.trip.2023.100852
Received 20 March 2023; Received in revised form 6 May 2023; Accepted 25 May 2023
Available online 31 May 2023
2590-1982/© 2023 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
A. Abulibdeh Transportation Research Interdisciplinary Perspectives 20 (2023) 100852
transport systems on mode choice in many cities around the world (e.g., this behavior after the operation of the new metro in Doha city.
(Birolini et al., 2019; Creemers et al., 2012; Golias, 2002; Hensher & Furthermore, this study aimed to understand the modal preferences,
Rose, 2007; Jou et al., 2011)). These studies aimed to determine the trips, and demographic characteristics of airport users. This was done by
factors that encourage air travelers to use the metro to commute to focusing on airport trips in Doha city to Hamad International Airport
airports. Some studies have found that these benefits are much smaller (HIA) to gain a better understanding of the factors shaping the airport air
than those expected by actors fostering the systems. Conversely, other travelers’ current mode choice and how the introduction of the new
studies found that some attributes such as travel time saving, travel cost, mode will change the current choice. Specifically, the objective of this
trip purpose, number of luggage, and user-friendly nature of the modal, study was to enhance our perception of the travelers’ requirements,
are significant in ground mode choice for air passengers (Jou et al., anticipate the potential impact of the new metro, and to identify the
2011). Some studies found that airport attractiveness is highly elevated driving forces that encourage travelers to use it as their primary choice
by an extension of the rapid-transit link and increases the dominance of to commute to the airport.
specific airports compared with others in the same area (Bergantino Previous studies on travel mode choice presented consistent and
et al., 2020). Furthermore, Bergantino et al. (2020) found that intro important insights into the variables that affect the travelers’ mode
ducing a new mode, such as a new express bus service, reduces the ac choice of the existing access mode. Most of these studies were conducted
cess time by 15 min (Tsamboulas & Nikoleris, 2008). in developed or developing countries where public transportation is an
The present analysis builds on previous studies that provide a important element of the transportation system. Unlike these studies,
background on different facets and explanations relevant to the subject the focus of this study was the use of public transportation in a devel
area and act as a starting point for the research. A review of the literature oping country that is currently working on introducing the metro as a
provided the following information: new mode of public transportation at the time this survey was con
First, previous studies on airport travel mode choice have catego ducted. Furthermore, this study aims to compare between two statistical
rized the explanatory variables into two groups to investigate the airport models and one machine-learning model in terms of their performance
travel mode choice (e.g., (Birolini et al., 2019; Pasha et al., 2020; Reza in predicting future mode choice.
Mamdoohi et al., 2012; Yaylali et al., 2016; Yazdanpanah & Hosseinlou,
2016)). One group consists of explanatory variables related to trip fea 2. Study area
tures (i.e., travel time and cost, number of travelers and their associated
luggage, travel time to the airport from trip origin, airplane departure Qatar is located in the far eastern part of the Arabian Peninsula and
times, out-of-vehicle and in-vehicle travel times, and trip purpose). The has a geographical area of 11,437 square kilometers (Balakrishnan et al.,
other group consists of the socioeconomic features of the commuters, 2023). Qatar is currently experiencing high rates of population growth
which include gender, profession, employment status, household in as a result of massive urban development, rising government expendi
come, age, and nationality. tures, and largescale investment projects (Al-Awadhi et al., 2022; Zaidan
Second, these studies are based on RP surveys conducted to amass & Abulibdeh, 2020). Doha city is the capital of Qatar and its most
data related to the explanatory variables to investigate the research populous city. The city has experienced significant economic growth and
question. In this type of survey, respondents indicated their mode of urban expansion over the past three decades (Abulibdeh and Zaidan,
choice to travel to the target destination from their origin and provide 2017). It is an urban primacy city in the Gulf region and is considered to
information about the trip characteristics (such as the number of be the hub of the state of Qatar in terms of economic development,
luggage items, trip cost, and time). They were also asked about their population, and culture. It also serves as the basis for regional devel
socioeconomic characteristics (such as household income, gender, age, opment. The location of the city has contributed to its becoming a
employment status, etc.) (Abulibdeh, 2020; Cirillo et al., 2014; Hasnine junction of transport routes (Abulibdeh, 2022; Ghofrani et al., 2022).
et al., 2019; Ouda et al., 2013; Petrik et al., 2016). The discovery of significant amounts of oil and gas, and population
Third, the collected data were analyzed using different discrete growth in cities, began to accelerate in the 1970 s, attracting a large
choice models (binary, mixed, and multinomial logit models). These number of expatriates to work in different fields. As a result, the urban
models are characterized by their simple probabilistic choice function, area of the city expanded by approximately 640% at an annual average
clarity in algebraic manipulations related to the derivation of the final rate of 19.6% between 1990 and 2000, compared with 8.9% between
probabilistic choice function, and ease of interpretation of the estima 2000 and 2017 (Abulibdeh et al., 2019a; Abulibdeh, 2019b) as shown in
tion results (Gokasar & Gunay, 2017; Gunay & Gokasar, 2021; Jiang Fig. 1.
et al., 2021; Abulibdeh et al., 2018; Pasha et al., 2020). Qatar is classified as an arid or semi-arid country with hot summers
Fourth, the literature presents compatible and key insights with and mild winters (Abulibdeh, 2021a, 2021b; Timothy, 2018).
respect to driving forces that influence the travelers’ choice of specific Commuting using bicycles or by walking is difficult and unadvisable,
travel mode to get to airports, particularly the time spent traveling and particularly during the summer season. Since its establishment in 1970,
the cost of the trip (Birolini et al., 2019; Jiang et al., 2021). Qatar has planned and built a transportation system around the move
Fifth, the latest computational progress has facilitated simpler ment of automobiles. In parallel with urban expansion, Doha city has
implementation of machine learning models for the examination of experienced a rapid increase in the number of private vehicles in use.
travel behavior. However, studies conducted in this area are yet to be Owning a car is a trademark of this country, where a high percentage of
comprehensive or decisive. Statistical analysis and machine learning households — particularly Qataris — own state-of-the-art cars. Qatari
models were rarely employed together in the preceding research to citizens consider owning more than one vehicle to be normal and a
scrutinize mode selection and contrast the outcomes of these models, source of pride. This has influenced the use of public transport networks
along with their efficiency in prognosticating mode choice. The pre currently and will continue to do so in the future (Mohammed et al.,
dominant advantage of many machine learning models lies in their lack 2023). Qatari and non-Qatari travelers in Qatar show a strong desire for
of stringent statistical assumptions, which allows for their adaptable private automobiles because of the lack of diverse transport options and
utilization across various data structures (Wang & Ross, 2018). inferior public transport services available in the city. Moreover, auto
Completion of the Doha Metro influences the tendency of the trav mobile ownership is often considered a status symbol that offers
elers to commute within the city and to the airport. Therefore, it was comfortable and safe travel opportunities for travelers. Automobile
deemed important to investigate the influence of the metro on mode ownership is the primary mode of choice in travel-related decision-
choice — taking the airport as a case study — to determine the effec making processes in Qatar.
tiveness of the metro in attracting new users. Therefore, this study aimed Public transportation was not popular in the country until it suc
to investigate the existing mode choice behavior and future changes in ceeded in getting selected to host the 2022 FIFA World Cup. The country
2
Fig. 1. Land-cover change and population growth in Doha (Abulibdeh et al., 2019).
spent and invested billions of dollars in developing its transportation origin, trip purpose, travel time to the airport, mode selection, and
infrastructure, including the public transportation network, to manage parking reimbursement (Abulibdeh and Zaidan, 2018; Earnhart, 2002;
the growth in air travel demand during this mega event, which will put Tseng et al., 2013). Conversely, the questions in the SP survey focused
large pressure on the only airport in the country to improve the capacity on hypothetical choice scenarios, such as using the Doha Metro in the
and level of service of access and egress of the ground travel mode future when it starts operating. These questions reflected what the
choice. The bus network was expanded with minimum fares ranging commuters said rather than what they actually did. Therefore, the SP
between QR 3 and 4 ($0.8–1) inside Doha and QR 4 – 9 ($1–2.5) outside data were used in this investigation to predict the future mode choice
the city (Mansour et al., 2022). However, this system is not popular in after Doha Metro will start operating and to evaluate the critical factors
the country and is used mainly by low-income expats, particularly those (e.g., trip and socioeconomic attributes) that will affect the individual
from India, the Philippines, Bangladesh, and other Asian countries. The decisions regarding the mode choice.
taxi system and Doha limousine service are options for travelers to One of the advantages of using the SP survey is that it allows us to
commute to airports and other places within the country; however, the investigate the possibility of using the metro to commute to the airport
number of operated taxis and limousines is still lower than the actual in a more comprehensive manner (Cherchye et al., 2015; Earnhart,
needs of the residents. Therefore, proposing a new public transportation 2002). Furthermore, the SP survey is more flexible than the RP survey
mode in a vehicle-oriented society requires thorough investigation of the (Cherchye et al., 2015; Earnhart, 2002; Tseng et al., 2013). However,
transport user preferences for private vehicle ownership and future user one of the main criticisms of this technique is the lack of reliability, as
preferences for a mode shift towards a new alternative. the expressed preference of the travelers may not coincide with the
Doha Metro is a state-of-the-art mass rapid transit system that is actual behavior because the answers are related to a hypothetical situ
being built in phases. Phase 1 includes building 37 stations along three ation. However, SP and RP self-administered questionnaire surveys have
main lines: red, green, and gold lines, and over an operational length of been used in many studies to collect primary data on ground access
approximately 76 km. The red line extends from south to north, while mode choices (Abulibdeh, 2018; Birolini et al., 2019; Ramsey et al.,
the green and gold lines extend from east to west in Doha city, as shown 2017). RP data are significant when analyzing the travel behavior of
in Fig. 2. existing transport alternatives using a discrete choice model. Therefore,
RP data were used to analyze the current mode choice (cars and taxi
3. Material and Methods services) of travelers to HIA. Models estimated using RP data have the
advantage of reflecting choices in real-world market settings. The results
3.1. Questionnaire development and data collection of the SP survey are significant for travel-demand forecasting for new
alternatives, as RP data cannot exist before new modes are implemented.
3.1.1. Survey design In this study, the questions were divided into three categories. The
An RP and SP survey was designed to gather information on trav first category was designed to gather information on travelers’ trip
elers’ travel behavior in choosing a mode choice to commute to HIA. SP characteristics. This part of the survey consisted of ten questions.
and RP self-administered questionnaire surveys were used to collect Travelers were asked to state the number of people traveling, number of
primary data. This method is suitable for investigating ground access luggage trips, trip purpose, time of departure, arrival time to the airport,
mode choices and enables us to reach a large number of people in a short and class (economy, business, or first). The second set of questions was
period. The questions in the RP questionnaire survey focused on the designed to gather information related to mode choice selection. This
current mode choice selection of air passengers, specifically their trip section consisted of 12 questions. Among these questions, respondents
3
Fig. 2. Qatar Metro lines and stations.
were asked to state their main mode of transport that they used to complete the survey. However, only 1247 of the surveyed households
commute to the airport, whether they were drivers or passengers, if they provided the complete information and were considered in the analysis.
always used the same mode to travel to the airport, how often they used
public transportation, parking reimbursement, factors influencing their 3.2. Methodology
choice of mode of transport, and their intention to use Doha Metro once
it begins to operate. The remaining questions were aimed at gathering The impact of Doha Metro as a new mode of choice was assessed by
information related to the socioeconomic characteristics of travelers, analyzing the socioeconomic and trip characteristics of the travelers in
such as income, age, gender, household size, nationality, and education. Doha using binary logistic regression, multinomial regression analysis,
and extreme gradient boosting (XGBoost). Furthermore, the SHapley
3.1.2. Data collection Additive exPlanations (SHAP) method was used to rank the input fea
To ensure a greater likelihood of a comprehensive response, the tures based on their importance according to the mean SHAP value. The
survey was randomly distributed only among air travelers who departed explanatory variables were classified into two categories: i) trip char
from the airport and waited in the boarding area, while connecting- acteristics, such as the number of luggage items and travelers, trip
flight air passengers were not included. The survey was conducted purpose, trip cost, and journey duration to the airport; and ii) socio
from January to March 2018 in both Arabic and English, as residents of economic attributes of commuters, such as household income, nation
Qatar are from different nationalities. The participants were selected ality, age, gender, education, and employment status. A detailed
based on a systematic approach: the first traveler sitting in the first row flowchart is shown in Fig. 3.
of the boarding waiting area was selected and given the survey; then, the In this study, airport travel mode choice was modeled according to
third traveler was selected, and so forth. If any traveler refused to the utility maximization principle combined with psychological choice
participate in the questionnaire, the next traveler was asked to partici behavior and the economic theory of consumer behavior. First, a chi-
pate. A total of 1546 air travelers were interviewed face-to-face to square test was used to examine the significant factors affecting the
4
Fig. 3. The research flowchart.
current mode choice, in particular, the significant differences in the equal to or greater than that associated with an alternative mode (j).
influence of trip characteristics and travelers’ socioeconomic attributes Therefore, the traveler selects the mode of transportation that yields the
on current and future mode choices. maximum utility.
To formulate a BL model, probability is expressed as in the following
3.2.1. Binary logit (BL) model Eq. (1):
The significant explanatory variables resulting from the chi-square
exp(βX1n ) 1 1
test were used to investigate the current airport travel mode choice Pn1 = = = (1)
exp(βX1n )+exp(βX2n) 1 + exp(βX2n − βX1n ) 1 + exp(ΔU)
using a BL regression model. This model has been used as a discrete
choice model for mode choice studies owing to its capability to predict where,
the possibility of occurrence of a specific event based on independent Pn1 is the probability that the traveler n selects the first mode.
variables (see for example, (Alhussein, 2011; Reza Mamdoohi et al., βXn1 is the utility function in which the traveler n selects the first
2012)). Utilizing this model deepens our perspective and understanding mode.
of the current mode choice in the country before introduction of the βXn2 is the utility function in which the traveler n selects the second
metro. The BL model was used to analyze the current mode choice and mode.
∑
predict a categorical variable from a set of predictor variables based on ΔU = βX2n − βX1n = (ai − bi )Zi , where Zi is the ith variable, ai is
the odds ratio between the variables. The dependent variable in the the coefficient of the i variable in βXn1, and bi is the coefficient of the ith
th
model was given a value of “0′′ for personal automobile use and ”1′′ for variable in βXn2.
taxi services. One of the advantages of using this model is that it controls
for potential determinants, including the traveler’s socioeconomic 3.2.2. Multinomial logit (MNL) model
characteristics and trip conditions. The MNL model was utilized to assess the impact of introducing the
In this study, the probability of selecting a specific mode (i) for Doha Metro as a mode of transportation to the airport. This model is
commuting was equal to the probability that the utility of mode (i) was based on the random utility theory. The concept underlying the model
5
analysis is that each alternative in the choice set provides the travelers SHAP is a coalitional game-theoretic method for describing the perfor
with some utility that can be expressed in terms of measurable or mance and output of any machine-learning model. It uses an additive
observable characteristics of both the traveler and alternative. The feature attribution method and establishes a link between the optimal
larger the difference in the utility between the two alternatives, the more credit allocation and local explanations using game theory’s traditional
likely the traveler is to choose the alternative with the higher utility. Shapley values and their related extensions. The SHAP method helps to
MNL can be expressed as explain different supervised learning models. Furthermore, this method
assigns an important value to each explanatory variable for a specific
e Ui
P(i) = ∑ Ui , (2) prediction (Mangalathu et al., 2020). Mathematically, SHAP is
j∊J e expressed as follows:
where, ∑
M
P(i) is the probability of a decision maker choosing alternative i; (5)
′ ′
g(x ) = ∅0 + ∅i xi
Ui and Uj are the utilities of alternatives i and j, respectively; and. i=1
J is the number of alternatives. where, g(x ) is the explanation model, x ∊{0,1}M i is the coalition
′ ′
vector or simplified features, M is the maximum coalition size, ∅i ∊ ℝ is

3.2.3. Extreme gradient boosting (XGBoost) the feature attribution for the feature i; the Shapley values × are the
In this study, the XGBoost (Chen & He, 2020) model was used to input variable in the model, and × = (x1, x2, …,xp), where p is the
classify the travel mode choices according to the travelers’ character number of input variables.
istics and trip condition variables. XGBoost is a framework of a gradient-
boosted decision tree-based ensemble method based on the idea of ad 3.2.5. Elasticity analysis
ditive training; it is used for both regression and classification, and is The final step was to perform the direct and cross-elasticity calcu
designed for speed and performance. In this model, each low-depth lations. In this study, elasticity represented the measures of socioeco
decision tree is built to minimize a defined loss function; however, nomic characteristics and trip condition variable sensitivity to changes
more weight is allocated to the cases that are incorrectly predicted by in mode choice. The aim of this step was to analyze the change in the
the previously developed trees. Therefore, the final XGBoost model re probability of choosing a specific mode choice for the current and future
sults are collectively determined by the results of all the developed trees. mode choices based on the change in the percentage of significant in
Furthermore, the XGBoost model enables an understanding of the sig dependent variables. The direct elasticity analysis was related to the
nificance of independent variables in explaining the dependent variable variables of the three transport modes under consideration (private car,
(Wang & Ross, 2018; Zaidan et al., 2022). The general unregularized taxi services, and metro). The cross-elasticity analysis was related to the
XGBoost algorithm is as follows. significant independent variables considered in the analysis (travelers’
For a given dataset, let xi = {xi1, xi2, …,xij} represent a vector of the socioeconomic characteristics and trip conditions). Direct elasticity and
observed values for the ith observation on j features, where i ∊ {1,2,3,…, cross-elasticity were computed using the following expressions (Larra
I} and j ∊ {1,2,3,…,J}, and ̂y i is the value of the target outcome for the ith naga et al., 2017; Ortúzar, 2011):
observation. Using K as the number of trees, a tree ensemble model is ( )
established to predict the output using additive functions, as shown in EPiq,Xikq = θik .Xikq 1 − Piq , (6)
Equation (3).
EPiq,Xjkq = − θjk .Xjkq .Pjq , (7)
∑
K
y i = Φ(xi ) =
̂ fk (xi ), fk ∊F , (3)
k=1
where EPiq,Xikq denotes the elasticity of the choice probability of the
alternative i for the individual q (Piq) of choosing the mode Ai, consid
where, ering a marginal change (1% increase) in a given variable Xikq with
F = space of regression tree {f(x) = wq(x)} (q: ℝJ → L, w ∊ ℝL) and fk respect to the base situation. EPiq,Xjkq denotes the elasticity of the prob
corresponds to an independent structure of each tree q that maps an ability of choosing the mode Ai, considering a marginal change in the
observation to the corresponding lth leaf and leaf weight w, where l ∊ value of the kth variable of the alternative Aj for the individual q, and θ is
{1,2,3,..,L). a constant.
A prediction output is obtained once q is developed and w is esti
mated. An observation of a leaf is first assigned to each tree to obtain the 4. Data analysis and results
output based on the values of the feature set and the sum of the weights
of the corresponding leaves. The following objective function is opti 4.1. Descriptive analysis
mized by the algorithm to develop several trees:
∑ ∑
L (Φ) = l(̂
y i , yi ) + Ω(fk ) (4) The entries in Table 1 show that the main modes of transportation
i k used by the air travelers to arrive at the airport were cars (82.2%),
followed by taxis (15.7%), limousines (a form of luxurious taxi) (2%),
where Ω(f) = γT + 12 λ‖w‖2 . Ω is a regularization term that penalizes and and buses (1%). For the purpose of analysis, these modes were grouped
controls the complexity of the model and smoothens the final learned into two categories because of the insufficient sample size. Taxis, lim
weights to prevent overfitting. l is the differentiable convex loss func ousines, and the other modes were combined and identified as taxi
tion. This function measures the difference between the observed yi and services, despite the fact that these modes may differ in some of their
predicted ̂y i . T is the number of leaves in the tree structure, while the characteristics. The percentage of passengers who commuted by bus to
term λ‖w‖2 denotes a form of L2 regularization on the leaf weights. For the airport was very low; hence, this mode was excluded from the model.
more information on XGBoost, readers can refer to Zopluoglu (2019) The high percentage of car ownership may explain the high percentage
and Zaidan et al., (2022). of users using this mode to travel to the airports. The table shows that
82.6% of the respondents owned one car or more. Another interesting
3.2.4. Shapley additive exPlanations (SHAP) observation is that 67.1% of the travelers used the same mode of
The SHAP method (Mangalathu et al., 2020) was employed in this transportation each time they commuted to the airport. This implies that
study for an in-depth analysis to rank the primary factors that influence cars have been the major mode of transportation for a long time.
the mode choice before and after the introduction of the Doha Metro. Furthermore, parking and parking fees are not considered obstacles
when using cars as the main mode choice. Although the majority of the
6
Table 1
Descriptive statistics of travelers’ trip conditions.
Airport access mode Number of bags Parking passengers’ cars Number of times the passengers traveled from
the HIA in the past 12 months
Car 78.6% 1 16.3% Airport long-term car park 16.7% 1 34.5%

Taxi 16.1% 2 21.4% Private long-term car park 11.7% 2 29.2%
Limousine 2.5% 3 14.3% Airport short-term car park 35.3% 3 18.4%
Bus 0.96% 4 14.5% Hotel 2.1% 4 8.7%
Others 1.84% 5 or more 33.5% others 34.2% 5 or more 9.2%
Primary factor influencing the choice of ground Using the same mode to Parking charges Purpose of the trip
access mode travel to HIA
Cost 12.4% Yes 68.8% Reimbursed in full 15.4% Holiday/leisure 63.7%
Journey time 26.5% No 31.2% Reimbursed partially 7.8% Visit relatives/friends 13.1%
Parking charges 14.9% None 76.8% Business 12.6%
Number of bags amount 10.8% Passengers traveling on Resident or visitor Others 10.6%
plane class
Public transportation availability 5.6% Economy class 81.8% Residents 95.1%
Nature of party 8.2% Business class 10.3% Visitors 4.9%
Others 21.6% First class 7.9%
travelers (80%) are not reimbursed when they use the parking facilities the ability of the classifier to correctly classify the unseen data points
at the airport, many of them still use these facilities owing to the low into different classes), precision, recall, and F1-score, which is the har
fees, which are QR 6 ($1.6) per hour for short-term and QR 45 ($12) per monic mean of precision and recall, were calculated across the ten cross-
day for long-term parking. Table 2 shows the socioeconomic charac validation folds.
teristics of travels that were contributed in this study. One challenge usually encountered in modeling travel mode choice
is the issue of unbalanced datasets. If this issue is present, the estimation
4.2. Statistical analysis verses machine learning modeling of the three models will be biased, which in turn will lead to a higher
prediction error for the classes of mode choice with smaller shares. The
In this study, three models were used to investigate and predict the majority of the trips are made by personal vehicles, whereas approxi
travel mode choices in Doha. To enhance the performance of the models mately 12.7%, 2%, and 2.1% of the trips are made by taxi, limousine,
in classifying unseen data and overcoming the overfitting and under and other modes, respectively, and only approximately 1% of the trips
fitting problems, a cross-validation method was applied. In this section, are made by bus. The dataset is unbalanced and may affect the accuracy
the RP data are used to compare the performance of the BL and XGBoost and performance of the models when predicting the mode choices with
models, while the SP data are used to compare the performance of the smaller shares, such as buses. Therefore, to reduce the imbalance effect,
MNL and XGBoost models. In K-fold cross-validation, the dataset gath taxi, limousine, and the other modes were combined into one set, and
ered by both RP and SP questions in the survey was split into K subsets of the bus mode was excluded from the study, as the number of travelers
equal size and then, the model was trained on all but one of the subsets who used this mode was very low.
and tested on the rest. This process was repeated until the model was The models were run 150 times to determine their average prediction
trained on each instance of the given data. The evaluation metrics were accuracy and robustness to data changes. The dataset was randomly
then computed and averaged across all the iterations. In this study, divided into training and testing subsets for each run. The training
cross-validation with ten-folds was conducted. The total errors of the subset encompassed 90% of the data, while the testing subset encom
two models and the classification error for each travel mode were passed the rest. The training and testing errors were averaged for the
averaged across the ten-fold range (Tables 3 and 4). Total error refers to 150 runs, as shown in Tables 3 and 4. The error was calculated based on
the number of misclassified trips out of the total number of trips. The the number of journeys predicted to have the wrong travel mode choice
mode-specific classification error was computed as the number of mis out of the total number of journeys. The three models illustrated good
classified trips out of the total number of trips made by the specified overall prediction accuracy for the complete choice set of two or three
mode. Furthermore, a set of well-known evaluation metrics was applied travel mode choices, with the XGBoost model having lower errors than
to evaluate the model’s classification performance (i.e., accuracy, pre the other models. Table 3 shows that the XGBoost model has total
cision, recall, and F1-score) (Tables 5 and 6). The mean accuracy (i.e., training and testing errors of 5.74% and 16.73%, respectively, when
Table 2
Descriptive statistics of travelers’ socioeconomic characteristics.
Age Monthly household income (QAR) Number of vehicles Employment status
18–24 18.7% Less than 5000* 6.8% None 15.1% Full-time worker 53.6%
25–34 29.4% 5,000–14,999 38.6% One car 29.8% Part-time worker 15.7%
35–44 36.8% 15,000–24,999 18.3% Two cars 28.3% Not employed 30.7%
45+ 15.1% 25,000 or more 36.3% Three cars or more 26.8%
Number of persons traveling Education Nationality Number of persons in household
1 14.3% Did not finish high school 1.4% Gulf Cooperation Council (GCC) including Qatari 30.3% 1 15.8%
2 22.5% Finished high school 5.3% North American and Europe 19.4% 2 16.7%
3 19.8% College 23.8% Arab (excluding GCC) 23.1% 3 15.7%
4 16.7% University 64.8% Asian 18.7% 4 22.5%
5 or more 26.7% Higher education 4.7% Others 8.5% 5 or more 29.3%
Gender Disability
Male 53.4% Yes 4.2%

Female 46.6% No 95.8%
*
$1 USA ¼ 3.64 QR.
7
Table 3
Average testing errors of BL and XGBoost models.
BL Model BL Model XGBoost Model XGBoost Model
Training Error Testing Error Training Error Testing Error
Mean Variance STD Mean Variance STD Mean Variance STD Mean Variance STD
Total 19.94% 0.001 0.0327 18.02% 0.0013 0.0364 5.74% 0.001 0.0471 16.73% 0.001 0.030
Private Car 12.02% 0.009 0.0315 12.13% 0.0013 0.0355 3.26% 0.005 0.0615 11.49% 0.001 0.037
Taxi services 28.12% 0.007 0.0822 28.50% 0.0073 0.0856 16.41% 0.015 0.0672 26.07% 0.002 0.047
Table 4
Average testing errors of MNL and XGBoost models.
MNL Model MNL Model XGBoost Model XGBoost Model
Training Error Testing Error Training Error Testing Error
Mean Variance STD Mean Variance STD Mean Variance STD Mean Variance STD
Total 17.84% 0.001 0.0327 19.68% 0.00031 0.0176 7.18% 0.0014 0.0471 12.46% 0.0023 0.0093
Private Car 2.92% 0.009 0.0315 4.06% 0.00174 0.0418 1.35% 0.0047 0.0615 2.76% 0.0002 0.0150
Taxi services 2.12% 0.007 0.0822 3.70% 0.00275 0.0525 1.61% 0.0148 0.0672 2.38% 0.0004 0.0187
Metro 32.76% 0.015 0.3280 34.27% 0.00335 0.0578 27.93% 0.0035 0.0498 32.26% 0.0012 0.0341
Table 5
Average of the evaluation metrics of the classification models BL vs. XGBoost.
BL Model XGBoost Model
Accuracy Precision Recall F1-score Accuracy Precision Recall F1-score
Total 80.893% 81.97% 81.98% 81.77% 83.26% 83.27% 83.15%

Private Car 84.75% 87.88% 86.20% 82.25% 85.84% 88.52% 87.12%
Taxi services 77.04% 71.50% 73.89% 78.67% 73.93% 76.09%
Table 6
Average of the evaluation metrics of the classification models MNL vs. XGBoost.
MNL Model XGBoost Model
Accuracy Precision Recall F1-score Accuracy Precision Recall F1-score
Total 80.32% 82.38% 80.32% 82.32% 87.54% 88.83% 87.53% 86.99%

Private Car 80.28% 85.93% 84.92% 83.07% 97.24% 89.58%
Public Transportion 84.59% 86.30% 89.49% 86.52% 97.62% 91.70%
(Taxi, limousine)
Metro 90.26% 63.72% 73.56% 96.89% 67.74% 79.68%
considering two different modes, whereas the model has a total training order of the answers to each question of the data instances is presumed
and testing error of 7.18% and 12.46%, respectively, when considering to behave as players in a coalition (Zaidan et al., 2022). Fig. 4 shows the
three mode choices (see Table 4). The amount of these errors is less than mapping of these answers to understand how these variables influence
that resulting from the other two models. Furthermore, the three models the mode choice. The importance factors of the input variables, shown in
have a high total rate of accuracy when considering either two or three Fig. 4, were calculated as the average of the absolute Shapley values per
modes, and XGBoost performs better than the other two. This applies to feature across the RP data. The input features were ranked based on
each mode where the prediction accuracy is high, as shown in Tables 5 their importance according to the mean SHAP value. The higher the
and 6. The BL and MNL models can replicate the shares of all the choices; mean SHAP value, the more important was the variable. The figure also
hence, the predicted mode shares resulting from using these two models shows the importance of each input variable for the mode choice, that is,
are similar to the observed ones, as shown in Tables 3 and 4. However, private cars and taxi services. Each point in the Figure represents the
the XGBoost model performed well in predicting the mode choice, and it Shapley value for the input variables. The y-axis represents the order of
showed its capability to lower the overall prediction error. Although the importance of the input variable from top to bottom. Furthermore, each
total accuracy percentage was high in the three models, it was affected point in the figure related to the input variable is colored by the value of
by unbalanced data. the input variable from low (blue) to high (red). The density of the
points indicates the distribution of the dots in the RP dataset. The
number of times the public transport is used (taxi services) is the most
4.3. Assessing current and future airport travel mode choice
important input variable in the mode choice. Fig. 4 shows that the
higher is the “times using public transport” value, the higher is its SHAP
The first step in assessing the current mode choice was to map the
value, and the larger is its impact on the mode choice selection. The
importance of the travelers’ socioeconomic characteristics as well as the
respondents were asked how often they used public transportation (taxi
trip condition variables on the mode choice based on the rank of the
services), and their answer choices are shown in Table 7. The figure
answers to the survey questions, as shown in Table 7. The impact of
shows that those who answered the first choices (i.e., daily/weekly)
socioeconomic characteristics and trip conditions on the mode choice
tended to use public transportation (blue color), while those who used it
can be further analyzed using the SHAP method. In this method, the
8
Table 7 other nationalities did. This is due to their high level of income and their
Some questions of the survey and their designated answers. preference for using personal automobiles in their commuting activities.
Questions Respondents’ answers choices Asians are less likely to use cars, perhaps because of their low incomes.
Furthermore, the figure shows that fewer travelers and younger trav
Times Using of public (1) Daily; (2) Weekly; (3) Monthly; (4) Annually; (5)
transportation Don’t use. elers tend to use public transportation. However, as the number of
Nationality (1) Qatari; (2) Gulf Cooperation Council (GCC) owned vehicles increases, travelers tend to use their cars as their main
citizens; (3) European; (4) North American; (5) Arab, mode choice. This figure only shows the tendency of the travelers to
(6) Asian; (7) Others choose their mode but does not show which factors are significant in
Number of persons in (1) 1; (2) 2; (3) 3; (4) 4; (5) 5 or more
household
choosing the mode.
Vehicle ownership (1) 0; (2) 1; (3) 2; (4) 3 or more Fig. 5 shows the impact of introducing the Doha Metro as a new
Employment status (1) Full-time; (2) Part-time; (3) Not employed mode choice for feature importance. The metro will take shares either
Number of travelers Stated by the interviewed person from those using cars or taxi services as the main modes of travel. The
Age (1) 18–24; (2) 25–34; (3) 35–44; (4) 45–54; (5) 55–64;
main socioeconomic features that encourage travelers to switch to using
(6) 65 or older
Reimbursed parking fees (1) Full reimbursement; (2) Partial reimbursement; (3) the metro include age, nationality, occupation, vehicle ownership, and
No reimbursement income. Class category and trip purpose are among the most important
Airport access mode (1) Car; (2) Taxi; (3) Limousine; (4) Bus; (5) Others trip condition variables that encourage travelers to use metro services.
Trip purpose (1) Vacation; (2) Visit family/friends; (3) Business; (4) The figure shows that the importance of the features underwent some
Others
Monthly household (1) Less than 5,000; (2) 5,000–9,999; (3)
changes. The “times using public transportation” is still the most
income (QAR) 10,000–14,999; (4) 15,000–19,999; (5) important feature in the mode choice. However, the Shapley value for
20,000–24,999; (6) 25,000–29,999; (7) the metro is very low compared with the use of private cars and taxi
30,000–34,999; (8) 35,000 or more services (public transportation). The number of persons in the household
Occupation (1) Official; (2) Manager/Specialist; (3) Sales/services;
is the second most important feature to determine the mode choice after
(3) Manufacturing; (4) Others
Using the same mode to (1) Yes; (2) No introducing the metro, whereas it was the person’s nationality prior to
travel to HIA that. Although these figures show the importance of the input variables
Number of luggage Stated by the interviewed person in mode choice selection, the SHAP method does not indicate whether
Gender (1) Male; (2) Female these features are significant in mode choice selection. Therefore, there
Class category (1) Economy class; (2) Business class; (3) First class
is a need for statistical analysis to determine the significant variables.
less often tended to use the car as the main mode choice of travel (red 5. Statistical analysis
color).
In terms of nationality, Qataris and Gulf Cooperation Council (GCC) To assess the current mode choice, we considered two alternative
citizens preferred to use private cars more than those belonging to the airport travel modes: personal automobiles and taxi services. A BL model
Fig. 4. Interpretation of features of binary classification.
9
Fig. 5. Interpretation of features of multiclass classification.
was used in modeling this current mode choice. However, MNL and
Table 8
XGBoost models were designed and used for modeling the airport travel
Results of Chi-square test on main variables influencing current airport travel
mode choice after introduction of the metro. The aim was to identify the
mode and future use of Doha Metro.
elements that might influence car users to shift to this mode. In the
Factors influencing mode choice Current mode Future mode
model, the dependent variable was given a value of “1′′ for cars, ”2′′ for
choice choice
taxi services, and “0′′ for metro use. The variables used were the same as (Personal vehicle (Personal vehicle
those used to assess the current mode choice using the BL model. vs. taxi services) vs. taxi services vs.
The explanatory variables identified in the survey were extensively Doha Metro)
evaluated to design an appropriate travel-mode choice model. These X2 value P X2 P
variables were assessed to determine and identify the significant vari value value value
ables that effectively augmented the data for the mode-choice model Trip characteristics
using the chi-square test. Some of the categories in these explanatory Number of people traveling 57.835 0.000 76.257 0.000
variables were combined to ensure reliability in conducting the chi- Number of bags 18.853 0.008 42.386 0.000
squared test by ensuring that each cell in the cross-tabulation had a Traveling class 42.491 0.000 14.926 0.008
Trip purpose 8.622 0.039 9.627 0.036
count of five or more. Place to park the car 93.738 0.000 4.217 0.269
The chi-square test was used to compare the socioeconomic factors Reimbursement of parking fees 48.249 0.000 8.470 0.009
and trip characteristics that were significant in selecting the current or Trip conditions (e.g., time) 85.545 0.000 24.546 0.000
future mode choice. Based on the results of the chi-squared test, all the Mode of transportation currently – – 0.276 0.763
used
socioeconomic factors and trip characteristics were determined to be
Socioeconomic characteristics
significant in explaining the current mode choice (Table 8). However, Nationality 184.825 0.000 16.239 0.012
the results of the chi-square test showed that finding a place to park a car Age 116.653 0.000 25.224 0.000
is not a significant trip characteristic feature in selecting the future mode Gender 64.194 0.000 0.827 0.485
choice. This also applies to gender as a socioeconomic factor, which Income 95.383 0.000 10.852 0.036
Vehicle ownership 204.639 0.000 34.898 0.000
affects the use of future mode choice. Based on the chi-square test re Employment status 137.270 0.000 8.539 0.026
sults, all the variables found to be insignificant were excluded from the Disability 10.482 0.000 34.629 0.000
model. Other variables that had no direct influence on the selection of a
specific mode were omitted from the model. One of these variables was
the reimbursement of parking fees because those who used taxi services cars), employment status, and part-time employment (see Table 9).
would not pay the parking fees. The number of respondents was Upon introduction of the metro, some of these factors became significant
considered a continuous variable, whereas the other variables were in using the car but not in using the taxi services mode, and vice versa.
treated as categorical variables. Furthermore, the importance of these factors changed, as shown in
A comparison of the driving factors influencing the mode choice Fig. 6. In that case, and in terms of using personal vehicles, the signifi
before and after implementation of the Doha Metro revealed that some cant factors affecting the negative mode choice were the flight class
of these factors were the same. The significant factors that negatively category (economy class), nationality (North American and Europe),
influenced the current mode choice were the number of travelers, full and car ownership. The significant factors affecting the positive mode
parking fee reimbursement, journey time, nationality, vehicle owner choice were parking fee reimbursement (full and partial), nationality,
ship (owning one or more cars), age (35 – 44 years), and full-time and age (Table 10). In terms of taxi services, the significant factors
employment. However, the significant factors that positively influ affecting the negative mode choice were the number of luggage trips,
enced the current mode choice included the number of luggage, age (18 journey time, age, car ownership (owning one car), and working part-
– 34 years), average household income, vehicle ownership (owning no time.
10
Table 9 most important socioeconomic features and trip characteristic factors in

Model parameter estimates for current mode choice (private car = base selecting the mode choice prior to the introduction of the metro,
category). respectively. However, age, number of persons in the household, and
β Std. Sig. Exp number of times that passengers used to use taxi services were the most
error (β) important socioeconomic features and trip characteristics in selecting
Number of travelers (5 travelers or the mode choice. Introducing Metro as a mode choice shows that the
more = reference category) variables that affect the mode choice include income as the most
1 traveler -0.841 0.465 0.023 0.737 important variable, followed by nationality and the number of pieces of
2 travelers -0.991 0.415 0.016 0.861
luggage, as shown in Fig. 6. This indicates that low-income travelers
3 travelers 0.593 0.393 0.131 1.809
4 travelers 0.617 0.381 0.105 1.853 may choose Metro during their trip to the airport.
Number of luggage (5 luggage or more
= reference category) 6. Trip characteristics
1 luggage -0.777 0.398 0.051 0.860
2 luggage − 1.351 0.357 0.000 0.659
3 luggage − 1.022 0.344 0.003 0.622
Some explanatory variables related to trip characteristics were found
4 luggage − 1.065 0.327 0.001 0.445 to be significant and influence the current and future mode choices as
Trip purpose (Business = reference shown in Table 9 and Table 10. The number of travelers in the different
category) groups was a significant explanatory variable influencing the current
Holiday/leisure -0.720 0.326 0.075 0.912
and future mode choices. The 5 or more travelers category is the refer
Visit relatives/friends − 1.373 0.431 0.012 0.510
Class category (First class = reference ence category for the various number of travelers categories. Compared
category) with this reference category, in the current mode choice, the coefficient
Economy class 0.814 0.284 0.009 1.598 (β) for the variable “number of travelers” was negative for 1 traveler and
Business class − 1.062 0.390 0.078 0.646
2 travelers categories (β = − 0.841, β = − 0.991). This implies that a
Reimbursed parking fees (not
reimbursed = reference category)
unit increase in the number of travelers who travel alone or travel in a
Reimbursed in full − 1.052 0.240 0.000 0.564 group of two decreases the likelihood of choosing taxi by 26.3% and
Reimbursed partially -0.781 0.319 0.044 0.784 13.9%, respectively. On contrast, the coefficient (β) is positive but not
Primary factor significant for the other categories of the “number of travelers” variable.
Cost − 1.280 0.493 0.000 0.455
Introducing the Metro attracts some travelers who used to use their cars
Journey time -0.550 0.235 0.019 0.577
Nationality (Others = reference or taxis. The entries in Table 10 shows that the coefficient is negative for
category) 1 and 2 travelers categories and positive for 3 and 4 travelers categories.
Qatari and GCC -0.959 0.346 0.000 0.526 The table shows that a unit increase in the number of travelers who
North American and Europe 0.366 0.280 0.191 1.042 travel alone or in a group of 2 decreases the likelihood of choosing cars
Arab (excluding GCC) -0.611 0.279 0.028 0.543
Asian 1.294 0.297 0.000 1.574
by 21.9% and 7.2% and of choosing taxi by 13.9% and 27.9%, respec
Age (45 years or older = reference tively. Conversely, a unit increase in the number of travelers who travel
category) in a group of 3 or group of 4 increases the likelihood of choosing cars by
18 – 24 years -0.789 0.387 0.128 0.955 28.1% and 42.3% and of choosing taxi by 21.5% and 42.6%,
25 – 34 years -0.935 0.262 0.066 0.874
respectively.
35 – 44 years − 1.201 0.245 0.001 0.670
Income (25,000 or more = reference Similarly, the number of luggage items was significant in deter
category) mining the current and future mode choices. The 5 or more luggage
Less than 5,000 1.253 0.421 0.003 3.500 category is the reference category for the various number of luggage
5,000 – 14,999 0.377 0.209 0.072 1.457 categories. The coefficient of this variable was negative in determining
15,000 – 24,999 -0.223 0.239 0.350 0.800
Vehicle ownership (Three vehicles or
the current mode choice for all categories as shown in Table 9. The
more = reference category) entries in the table shows that a unit increase in the number of luggage
One Vehicle − 1.371 0.367 0.000 0.839 decreases the likelihood of choosing taxi by 14%, 24.1%, 37.8%, and
Two vehicles − 1.599 0.262 0.000 0.549 55.5% for those who travel with 1, 2, 3, and 4 luggage, respectively.
Employment status (Not employed =
Introducing the Metro has no significant effect of travelers who travel
reference category)
Full-time worker -0.766 0.346 0.005 0.628 with 1 luggage using either their cars or a taxi. On the contrary, traveling
Part-time worker 0.952 0.406 0.075 1.386 with more than one luggage increases the probability of choosing either
a car or a taxi for traveling at the expense of the metro in the future. The
*Cox and Snell R Square: 0.411, Nagelkerke R Square: 0.526.
The literature suggests that values of 0.2 to 0.4 for R2 represent an excellent fit
entries in the table shows that a unit increase of carrying 3 or 4 luggage
(Abulibdeh, 2018). increase the likelihood of choosing car by 25.3% and 84.1% and taxi by
28.4% and 42.8%, respectively.
Trip purpose is an important and significant variable on determining
Table 10 shows the parameter estimates that indicate the influence of
the current and future mode choices. The business category was
each factor for travelers’ mode choice of Metro relative to private car
considered as the reference category for trip purpose. In the current
and taxi services. The coefficient (β) values are the estimated MNL
mode choice, the coefficient (β) was negative for traveling for the pur
regression for the models. The negative values of the coefficient indicate
poses of holiday, leisure, and visiting relatives or friends. Table 9 shows
that the factors decrease the likelihood of that response category with
that a unit increase of travelling for holidays or visiting relatives de
respect to the reference category. The table also shows the Exp(β) values,
creases the likelihood of choosing taxi by 8.8% and 49% for traveling for
which represents the odds ratio for each category of the predictors.
spending a holiday and visiting relatives, respectively. In contrast,
Fig. 6 ranks the importance of the socioeconomic factors and trip
introducing the Metro will not encourage travelers to use it in their
characteristics in selecting the mode choice prior to and after intro
commuting. A unit increase of travelling for holidays or visiting relatives
duction of the metro using the F-score. The XGBoost model consists of a
increases the likelihood of choosing a car by 136.9% and 69.3% and
number of boosted trees that represent the estimators. Each feature is
choosing a taxi by 35.1% and 23% for traveling for holiday or visit
represented by a node in the tree and the number of nodes split to make
relatives, respectively.
the final decision represents the F-score. The figure clearly shows that
The flight class category was another variable related to trip char
nationality and number of times the public transportation is used are the
acteristics that was significant and had influence on the current mode
11
Fig. 6. F-score of feature importance: (a) prior to introduction of the metro, and (b) after the introduction of the metro.
Table 10
Model parameter estimates for using the Doha Metro, future mode choice, (Metro is the reference).
Cars* Taxi services*
β S.E. Sig. Exp(β) (odds ratio) β S.E. Sig. Exp(β) (odds ratio)
Number of travelers (5 travelers or more = reference category)

1 traveler -0.867 0.275 0.012 0.781 -0.936 0.326 0.083 0.861
2 travelers -0.558 0.286 0.047 0.928 -0.852 0.241 0.062 0.721
3 travelers 0.747 0.670 0.009 1.281 0.639 0.451 0.023 1.215
4 travelers 1.153 0.266 0.004 1.423 0.955 0.379 0.011 1.426
Number of luggage (5 luggage or more = reference category)
1 luggage -0.753 0.266 0.184 0.923 -0.641 0.236 0.213 0.741
2 luggage 0.958 0.286 0.021 1.148 -0.859 0.341 0.113 0.852
3 luggage 1.126 0.225 0.015 1.253 0.798 0.459 0.036 1.284
4 luggage 1.440 0.306 0.002 1.841 1.153 0.396 0.004 1.428
Class category (First class = reference category)
Economy class − 1.136 0.368 0.007 0.695 − 0.720 0.321 0.019 0.526
Business class 1.366 0.456 0.004 1.572 1.643 0.428 0.000 1.736
Trip purpose (Business = reference category)
Holiday/leisure 1.237 0.257 0.003 2.369 0.993 0.461 0.015 1.351
Visit relatives/friends 0.926 0.361 0.034 1.693 0.837 0.285 0.003 1.23
Reimbursed parking fees (not reimbursed = reference category)
Reimbursed in full 0.510 0.248 0.005 1.948 – – – –
Partially 0.461 0.313 0.094 1.631 – – – –
Primary factor
Cost of the trip -0.735 0.368 0.012 0.564 -0.604 0.361 0.021 0.753
Journey time − 1.152 0.274 0.003 0.726 − 0.497 0.260 0.036 0.608
Nationality (Others = reference category)
Qatari and GCC 0.867 0.275 0.002 1.381 − 0.787 0.338 0.034 0.629
North American and Europe 0.757 0.308 0.005 1.124 − 0.899 0.295 0.036 0.805
Arab (excluding GCC) 0.692 0.279 0.018 1.096 1.211 0.436 0.006 1.250
Asian − 1.169 0.293 0.044 0.584 1.357 0.329 0.002 1.393
Age (45 years or older = reference category)
18 – 24 years -0.807 0.248 0.080 0.793 0.694 0.358 0.042 1.152
25 – 34 years 0.864 0.373 0.014 1.166 -0.874 0.438 0.032 0.891
35 – 44 years 1.294 0.243 0.033 1.516 − 0.658 0.280 0.019 0.518
Income (25,000 or more = reference category)
Less than 5,000 -0.751 0.358 0.081 0.673 0.692 0.362 0.008 1.352
5,000 – 14,999 -0.560 0.200 0.057 0.942 0.854 0.239 0.036 1.138
15,000 – 24,999 1.281 0.217 0.007 1.324 0.546 0.451 0.002 0.731
Vehicle ownership (Three vehicles or more = reference category
One Vehicle 0.792 0.250 0.002 1.453 -0.672 0.269 0.026 0.881
Two vehicles 1.544 0.250 0.000 2.580 − 0.874 0.422 0.000 0.460
Employment status (Not employed = reference category)
Full time 1.197 0.365 0.022 1.682 -0.873 0.337 0.005 0.840
Part time -0.626 0.247 0.089 0.732 -0.632 0.469 0.076 0.535
* Reference is the Metro
* Cox and Snell R Square: 0.453, Nagelkerke R Square: 0.518, McFadden: 0.184.
12
choice and on the future mode choice. Regarding the current mode, the by 57.4% and by introducing the Metro this probability decreases to
coefficient (β) is positive for economy class and negative for business 39.3%, while the probability of choosing a car decreases by 41.6%.
class. This implies that a unit increase in traveling on economy class Age was a significant explanatory variable in explaining the current
increases the likelihood of choosing taxi by 59.8%, and decreases the travel mode choice, but it also had influence on the future travel mode
likelihood of choosing taxi by 35.4% when traveling on business class. choice. For the current mode choice, as age increased, the probability of
Introducing the Metro decreases the likelihood of using the car and taxi selecting taxi services decreased (Table 9). The 45 years or older age
when traveling on the economy class and increases the likelihood of category is the reference category for the various age categories. Air
using the car and taxi when traveling on business class. passengers aged 25–44 years old are less willing to choose tax over cars.
A unit increase in travelers in 25–34 age group decreases the likelihood
6.1. Reimbursement of parking fees of choosing taxi by 12.6%, while a unit increase in 35–44 age group
decreases the likelihood of choosing taxi by 33%. When assessing the
Another significant factor that influences the current mode choice is future mode choice, the 35–44 year age cohort variable was significant.
the reimbursement of parking fees. The coefficient of full reimbursement However, air passengers aged 35–44 years preferred to switch from
for the parking fees variable had a negative sign prior to introduction of using taxi services to using the metro or their private cars more than the
the Doha Metro and a positive sign after its introduction. The no reim other age groups (Table 10).
bursement category is the reference category for the various reimbursed Household monthly income was another significant indicator of the
parking fees categories. Prior to the introduction of the metro, the results current airport travel mode choice and had significant influence on the
show that full reimbursement for the parking fees variable has an odd future choices. In terms of the current situation, as income increases, the
ratio of − 1.052, meaning that a unit increase in reimbursing the parking probability of selecting personal automobiles increases. Low-income
fees decreases the likelihood of choosing taxi by 43.6%. After the travelers (Income < QR 5000; 1 USD = 3.68 QR) were more willing to
introduction of the metro, full parking fee reimbursement was still sig use taxi services than high-income travelers (Income > QR 25,000), with
nificant, and the coefficient was positive with an odds ratio of 1.948 for an odds ratio of 1.781. As household income increases, the tendency to
reimbursed in full and 1.631 for partial reimbursement. This means that use taxi services decreases. For example, the coefficient of the air pas
a unit increase in full or partial parking reimbursement increases the sengers with an average monthly income between QR 15,000 and QR
likelihood of using the car by 94.8% and 63.1%, respectively. These 24,000 was negative (β = − 0.223), with an odds ratio of 0.738,
results indicate the importance of this variable in selecting the travel meaning that a unit increase in being within this income category de
mode choice. Although full parking fee reimbursement is significant creases the probability of choosing taxi services by 26.2%. The same
before and after the introduction of the metro, this feature is less pattern is found when introducing the Metro whereby as income
important than the other features, as shown in Fig. 6. decreased, the tendency to use personal vehicles decreased.
In a wealthy developing country such as Qatar, owning more than
6.2. Trip journey cost and time factors one vehicle is normal. This may have influenced the current and future
use of taxi services. In assessing the influence of the car ownership
Journey cost and time was a significant factor in determining the variable on the airport mode choice, the results indicated that this
current mode choice. An increase in journey cost and time results in a variable was significant in selecting the current mode (see Table 9). As
decrease in public transportation usage. The coefficient of the journey the number of vehicles per household increases, the tendency to use taxi
cost and time variables was negative with an odds ratio of − 1.280 and services decreases. For example, the sign of the coefficient of owning one
-0.550, respectively. This indicates that a unit increase in the cost and vehicle was negative, which meant that a unit increase of owing one
time decreases the likelihood of choosing taxi by 54.5% and 42.3%, vehicle decreases the likelihood of choosing taxi by 16.1%. Vehicle
respectively. The journey cost and time variables were also significant in ownership is still a significant factor in selecting future airport travel
influencing the future airport mode choices, including metro. Con mode choices. Introducing the Metro will influence the taxi services
cerning personal vehicle usage, the impact of journey cost and time was share of mode choice. For example, a unit increase in owing two cars
the same as that of the current airport mode choice. decreases the probability of choosing taxi by 54%, where the in the
current mode choice owing two cars decreases the probability of
6.3. Socioeconomic characteristics choosing taxi by 45.1%. This implies that some travelers who used to
select taxi to travel to the airport will choose the Metro once it is
Personal socioeconomic characteristics also influence the current introduced.
and future airport travel mode choices. Nationality was one factor that Employment status was another significant variable (P = 0.05 and P
significantly affected current and future airport mode choices. Qatari = 0.1) for the current airport travel mode choice and for the future
and GCC nationals currently use private automobiles to travel to airports choice after Doha Metro came into service. In terms of current mode
more than those belonging to other nationalities do. Being a Qatari or choice, full-time employees were less likely to use taxi services, as the
GCC resident reduces the probability of using taxi services in the present coefficient of this variable was negative, while part-time employees
analysis. The entries in Table 9 shows that a unit increase of being Qatari preferred to use taxi services, with an odds ratio of 1.386. However, the
or from the GCC countries decreases the probability of choosing taxi by availability of the metro encourages part-time employees to switch to
47.4%, while being from Arab nationalities decreases the likelihood of using this new mode to travel to the airport.
choosing taxi by 15.7%. On the contrary, being from East or Southeast
Asia increases the likelihood of choosing taxi by 57.4%. Thus, having the 6.4. Elasticity analysis
metro available will not encourage Qataris and GCC residents to switch
from using private automobiles. For these nationalities (Qataris and Direct and cross-elasticity analyses were performed to detect and
GCC), the odds ratios are 1.381 and 0.629 for using cars and taxi ser understand the changes in the probabilities of selecting a specific
vices, respectively. This means that a unit increase in being Qatar or transport mode choice based on the occurrence of percentage changes in
from the GCC countries increases the probability of choosing a car by the independent variables. Table 11 shows the results of the elasticity of
38.1% and decreases the probability of choosing a taxi by 37.1%. The mode choice probability calculations based on the significant variables
same trend applies for the North American and Europe nationalities. presented in Table 10, excluding full and partial parking re
Conversely, Asians are use taxis more than private cars and they are imbursements, because these variables are not applicable when using
more willing to use the Metro. In terms of the current mode choice, a taxi services or the metro. The direct and cross-elasticity values in
unit increase of being Asian increases the probability of choosing a taxi Table 11 show that travelers using private cars are highly sensitive to the
13
Table 11 consistency between the training and testing errors because of its
Elasticity analysis. capability to avoid overfitting. The XGBoost and MNL models explain
Variable Private Taxi Metro the relationships between the independent variables and travel mode
car services choices. Different factors were found to influence the current mode
Number of luggage Private car 1.54 − 0.39 − 0.28 choice, and many of these factors continued to influence future mode
Taxi − 0.14 0.94 − 0.28 choice after introduction of the Doha Metro. Furthermore, this study is
services among the studies that use statistical analysis and machine-learning
Metro − 0.14 − 0.39 1.32 techniques to investigate travel mode choice, and it presents a rela
Economy class Private car − 0.22 0.63 0.47
Taxi 0.16 − 0.15 0.47
tively comprehensive range of independent variables that are ready for
services practical use.
Metro 0.16 0.63 − 0.61 Analysis of the driving forces influencing airport travel mode choice
Journey time Private car 3.79 − 2.13 − 1.18 before and after introduction of the Doha Metro revealed many common
Taxi − 3.39 1.81 − 1.18
forces. The significant factors influencing both the current and future
services
Metro − 3.39 − 2.13 3.37 mode choices were the number of travelers, number of bags, trip pur
Qatari and GCC Private car 2.83 − 1.36 − 1.53 pose, flight class category, parking fees, full parking fee reimbursement,
Taxi − 1.47 1.88 − 1.53 cost of the trip, journey time, nationality, age, average household in
services come, vehicle ownership, and employment status. Personal automobiles
Metro − 1.47 − 1.36 1.76
North American and Private car − 0.23 0.47 0.52
were used extensively by travelers who traveled to airports more
Europe Taxi 0.13 − 0.64 0.52 frequently and for different trip purposes. However, other modes of
services transportation were used by those traveling for holidays or leisure
Metro 0.13 0.47 − 0.64 purposes, and hence, they will be more interested in using the metro in
35 – 44 years Private car 0.59 − 0.26 − 0.36
the future. Another key factor in determining the travel mode choice to
Taxi − 0.35 0.48 − 0.36
services the airport was parking charges — mainly long-term parking charges —
Metro − 0.35 − 0.26 0.58 where many travelers were unwilling to take their personal car and park
1 car Private car − 0.65 0.12 0.15 it at the airport and therefore, preferred to use other modes to get to the
Taxi 0.23 − 0.35 0.15 airport. However, parking charge reimbursement was a key factor in
services
encouraging the commuters to use their personal vehicles and park them
Metro 0.23 0.12 − 2.21
2 cars Private car 3.45 − 2.27 − 1.34 at the airport. Travel time was another significant factor that influenced
Taxi − 1.42 2.49 − 1.34 mode choice. The results revealed that commuting time influenced
services approximately 23% of the commuters to use their personal automobiles
Metro − 1.42 − 2.27 0.63
and avoid using other modes, particularly buses. These findings can help
Part time Private car 0.21 − 0.16 − 0.13
Taxi − 0.09 0.33 − 0.13 public authorities to develop transportation-related policy measures. Of
services course, the policies adopted by the Ministry of Transportation in Qatar
Metro − 0.09 − 0.16 0.25 must be consistent with other facets of the government transportation
policy. However, the findings in this study can help the Ministry design
incentives to encourage travelers to use the Doha Metro when it comes
number of pieces of luggage, the duration of the trip to the airport,
into operation and discourage the use of private cars. The models used in
nationality, and car ownership (owning two cars). These variables
this study show high prediction accuracy for travel mode choices in
appear to be the most important in the mode of choice to travel to the
Doha city before and after the introduction of the new metro. The per
airport. For example, an increment of 1% in the time of the trip to the
formance of the XGBoost model substantially exceeds that of both the BL
airport represents a 3.79% higher market share for those who use pri
and MNL models in predicting and improving the accuracy of predicting
vate cars. Conversely, a 1% reduction in the trip time to the airport
the mode choice.
should increase the demand for the other alternatives by 3.39%.
Future work should take place after the metro becomes fully opera
Changes in the remaining variables affect the probability of selecting the
tional and real data on metro ridership are gathered. Furthermore, it is
corresponding mode to a lesser extent. The elasticity results represent an
necessary to investigate metro accessibility for both residents and visi
important tool for analyzing demand response and model competition
tors to the country. Once operated, it is recommended that the author
when introducing a new mode of transportation. The elasticity analysis
ities undertake measures to augment the metropolitan metro system to
in this study allows transport planners to assess the demand response to
cover additional areas within the country, predominantly residential
the different mode choices considered in this study.
areas, with the aim of enhancing commuter convenience and mobility.
Such measures may include intensifying the frequency of metro services,
7. Conclusion and policy implications
particularly during peak hours, to minimize wait times for travelers.
Strategies and policies that are deemed important for improving con
Modeling travel mode choice is a dynamic and significant step in
nectivity with other modes of public transportation are important and
travel demand forecasting in Doha city due to the transport infrastruc
should be pursued in future research. Investigating these factors will
ture development in preparation for the FIFA 2022 World Cup and the
contribute to the success of metro attractiveness.
introduction of the metro as a new mode of transportation. In this study,
three models were used to investigate and predict travel mode choices in
CRediT authorship contribution statement
Doha. These models are the extreme gradient boosting (XGBoost), bi
nary logit (BL), and multinomial logit (MNL) models. The results of these
Ammar Abulibdeh: Conceptualization, Data curation, Formal
models were examined by comparing their average multiclass prediction
analysis, Investigation, Methodology, Project administration, Resources,
errors. The factors affecting the travel mode choice were categorized
Validation, Visualization, Writing – original draft, Writing – review &
into two groups: travelers’ characteristics and trip conditions. A set of
editing.
independent variables were selected when developing the three models
using the entire dataset. These variables were selected by developing a
Declaration of Competing Interest
chi-square test to determine the statistically-significant variables to
improve the goodness-of-fit of the models. MNL maintained a high
The authors declare that they have no known competing financial
14
interests or personal relationships that could have appeared to influence 1942787513Y.0000000008, 6(1), 23–35. https://doi.org/10.1179/
1942787513Y.0000000008.
the work reported in this paper.
Creemers, L., Cools, M., Tormans, H., Lateur, P.-J., Janssens, D., & Wets, G. (2012).
Identifying the Determinants of Light Rail Mode Choice for Medium- and Long-
Data availability Distance Trips: Results from a Stated Preference Study. Https://Doi.Org/10.3141/
2275-04, 2275, 30–38. https://doi.org/10.3141/2275-04.
Earnhart, D., 2002. Combining revealed and stated data to examine housing decisions
The authors do not have permission to share data. using discrete choice analysis. J. Urban Econ. 51 (1), 143–169. https://doi.org/
10.1006/JUEC.2001.2241.
Ghofrani, A., Zaidan, E., Abulibdeh, A., 2022. Simulation and impact analysis of
Acknowledgement behavioral and socioeconomic dimensions of energy consumption. Energy 240,
122502. https://doi.org/10.1016/J.ENERGY.2021.122502.
Open Access funding provided by the Qatar National Library. Gokasar, I., Gunay, G., 2017. Mode choice behavior modeling of ground access to
airports: a case study in Istanbul, Turkey. J. Air Transp. Manag. 59, 1–7. https://doi.
org/10.1016/J.JAIRTRAMAN.2016.11.003.
References Golias, J.C., 2002. Analysis of traffic corridor impacts from the introduction of the new
Athens Metro system. J. Transp. Geogr. 10 (2), 91–97. https://doi.org/10.1016/
S0966-6923(01)00033-3.
Abulibdeh, A., 2018. Implementing congestion pricing policies in a MENA region city:
Gunay, G., Gokasar, I., 2021. Market segmentation analysis for airport access mode
analysis of the impact on travel behaviour and equity. Cities 74, 196–207. https://
choice modeling with mixed logit. J. Air Transp. Manag. 91, 102001 https://doi.org/
doi.org/10.1016/j.cities.2017.12.003.
10.1016/J.JAIRTRAMAN.2020.102001.
Abulibdeh, A., 2021a. Analysis of urban heat island characteristics and mitigation
Halse, A. H., Østli, V., & Killi, M. (2019). Revealed and stated preferences for reliable
strategies for eight arid and semi-arid gulf region cities. Environ. Earth Sci. 80 (7),
commuter rail in Norway. Https://Doi.Org/10.1080/19427867.2019.1586088, 12
259. https://doi.org/10.1007/s12665-021-09540-7.
(4), 288–292. https://doi.org/10.1080/19427867.2019.1586088.
Abulibdeh, A., 2021b. Spatiotemporal analysis of water-electricity consumption in the
Hasnine, M. S., Rashedi, Z., & Habib, K. N. (2019). Making use of elicited confidence
context of the COVID-19 pandemic across six socioeconomic sectors in Doha City,
ratings of the SP survey in a joint RP-SP model: a case study of commuting mode
Qatar. Appl. Energy 304, 117864. https://doi.org/10.1016/J.
choices for TDM evaluations. Https://Doi.Org/10.1080/19427867.2019.1585687,
APENERGY.2021.117864.
12(4), 274–287. https://doi.org/10.1080/19427867.2019.1585687.
Abulibdeh, A., 2022. Time series analysis of environmental quality in the state of Qatar.
Hawas, Y.E., Hassan, M.N., Abulibdeh, A., 2016. A multi-criteria approach of assessing
Energy Policy 168, 113089. https://doi.org/10.1016/J.ENPOL.2022.113089.
public transport accessibility at a strategic level. J. Transp. Geogr. 57, 19–34.
Abulibdeh, A., Andrey, J., Melnik, M., 2015a. Insights into the fairness of cordon pricing
https://doi.org/10.1016/j.jtrangeo.2016.09.011.
based on origin-destination data. J. Transp. Geogr. 49, 61–67. https://doi.org/
Heinen, E., Harshfield, A., Panter, J., Mackett, R., Ogilvie, D., 2017. Does exposure to
10.1016/j.jtrangeo.2015.10.014.
new transport infrastructure result in modal shifts? Patterns of change in commute
Abulibdeh, A., Al-Awadhi, T., Al-Barwani, M., 2019a. Comparative analysis of the
mode choices in a four-year quasi-experimental cohort study. J. Transp. Health 6,
driving forces and spatiotemporal patterns of urbanisation in Muscat, Doha, and
396–410. https://doi.org/10.1016/J.JTH.2017.07.009.
Dubai. Dev. Pract. 29 (5), 606–618. https://doi.org/10.1080/
Hensher, D.A., Rose, J.M., 2007. Development of commuter and non-commuter mode
09614524.2019.1598335.
choice models for the assessment of new public transport infrastructure projects: a
Abulibdeh, A.O., Zaidan, E.A., Alkaabi, K.A., 2018. Empirical analysis of the
case study. Transp. Res. A Policy Pract. 41 (5), 428–443. https://doi.org/10.1016/J.
implementation of cordon pricing: Potential impacts on travel behaviour and policy
TRA.2006.09.006.
implications. Transport. Res. F: Traffic Psychol. Behav. 53, 130–142. https://doi.
Ibrahim, A.N.H., Borhan, M.N., M.d. Yusoff, N.I., Ismail, A., 2020. Rail-based public
org/10.1016/j.trf.2018.01.006.
transport service quality and user satisfaction – a literature review. Promet - Traffic
Abulibdeh, A., Zaidan, E., 2018. Analysis of factors affecting willingness to pay for high-
& Transportation 32 (3), 423–435. https://doi.org/10.7307/PTT.V32I3.3270.
occupancy-toll lanes: Results from stated-preference survey of travelers. J. Transp.
Jiang, Y., Yu, S., Guan, W., Gao, S., Feng, T., 2021. Ground access behaviour of air-rail
Geogr. 66, 91–105. https://doi.org/10.1016/j.jtrangeo.2017.11.015.
passengers: a case study of Dalian ARIS. Travel Behav. Soc. 24, 152–163. https://doi.
Abulibdeh, A., Zaidan, E., Abuelgasim, A., 2015b. Urban form and travel behavior as
org/10.1016/J.TBS.2021.03.006.
tools to assess sustainable transportation in the greater Toronto area. J. Urban Plann.
Jou, R.C., Hensher, D.A., Hsu, T.L., 2011. Airport ground access mode choice behavior
Dev. 141 (3) https://doi.org/10.1061/(ASCE)UP.1943-5444.0000227.
after the introduction of a new mode: a case study of Taoyuan International Airport
Abulibdeh, A., Zaidan, E., 2017. Empirical analysis of the cross-cultural information
in Taiwan. Transportation Research Part E: Logistics and Transportation Review 47
searching and travel behavior of business travelers: a case study of MICE travelers to
(3), 371–381. https://doi.org/10.1016/J.TRE.2010.11.008.
Qatar in the Middle East. Appl. Geogr. 85, 152–162. https://doi.org/10.1016/j.
Larranaga, A.M., Arellana, J., Senna, L.A., 2017. Encouraging intermodality: a stated
apgeog.2017.06.001.
preference analysis of freight mode choice in Rio Grande do Sul. Transp. Res. A
AbuLibdeh, A. (2017). Traffic Congestion Pricing: Methodologies and Equity
Policy Pract. 102, 202–211. https://doi.org/10.1016/J.TRA.2016.10.028.
Implications. Urban Transport Systems. https://doi.org/10.5772/66569.
Li, T., Guan, H., Ma, J., Zhang, G., & Liang, K. (2017). Modeling travel mode choice
Abulibdeh, A. (2019). Water-energy nexus challenges and opportunities in Qatar. In
behavior with bounded rationality using Markov Logic Networks. Https://Doi.Org/
Qatar: Political, Economic and Social Issues (pp. 209–227). nova. https://
10.1080/19427867.2017.1342945, 11(6), 303–310. https://doi.org/10.1080/
novapublishers.com/shop/qatar-political-economic-and-social-issues/.
19427867.2017.1342945.
Abulibdeh, Ammar. (2020). Planning for congestion pricing policies in the middle east:
Liu, S., Yao, E., Cheng, X., Zhang, Y., 2017. Evaluating the impact of new lines on
public acceptability and revenue distribution. Transportation Letters. https://doi.
entrance/exit passenger flow of adjacent existing stations in urban rail transit
org/10.1080/19427867.2020.1857908.
system. Transp. Res. Procedia 25, 2625–2638. https://doi.org/10.1016/J.
Al-Awadhi, T., Abulibdeh, A., Al-Masri, A.N., Bin Touq, A., Al-Barawni, M., El
TRPRO.2017.05.316.
Kenawy, A.M., 2022. Spatial and temporal changes in electricity demand regulatory
Mohammed, S., Alkhereibi, A.H., Abulibdeh, A., Jawarneh, R.N., Balakrishnan, P., 2023.
during pandemic periods: The case of COVID-19 in Doha, Qatar. Energy Strategy
GIS-based spatiotemporal analysis for road traffic crashes; in support of sustainable
Reviews 41, 100826. https://doi.org/10.1016/J.ESR.2022.100826.
transportation Planning. Transp. Res. Interdiscip. Perspect. 20, 100836.
Alhussein, S.N., 2011. Analysis of ground access modes choice King Khaled International
Mamdoohi, A.R., Saffarzade, M., Taherpour, A., Yazdan Panah, M., 2012. Modeling air
Airport, Riyadh, Saudi Arabia. J. Transp. Geogr. 19 (6), 1361–1367. https://doi.org/
passengers’ ground access mode choice a case study of IKIA feasibility study of
10.1016/J.JTRANGEO.2011.07.007.
congestion pricing of current tehran odd and even zone view project. Article in
Balakrishnan, P., Abulibdeh, A., Abul Kasem Kabir, T., 2023. Assessment of the impact of
International Journal of Modeling and Optimization 147–152. https://doi.org/
anthropogenic evolution and natural processes on shoreline dynamics using multi-
10.7763/IJMO.2012.V2.101.
temporal satellite images and statistical analysis. Water 15 (8), 1440. https://doi.
Mangalathu, S., Hwang, S.-H., Jeon, J.-S., 2020. Failure mode and effects analysis of RC
org/10.3390/W15081440.
members based on machine-learning-based SHapley Additive exPlanations (SHAP)
Bergantino, A.S., Capurso, M., Hess, S., 2020. Modelling regional accessibility to airports
approach. Eng. Struct. 219 https://doi.org/10.1016/J.ENGSTRUCT.2020.110927.
using discrete choice models: An application to a system of regional airports. Transp.
Mansour, S., Alahmadi, M., Abulibdeh, A., 2022. Spatial assessment of audience
Res. A Policy Pract. 132, 855–871. https://doi.org/10.1016/J.TRA.2019.12.012.
accessibility to historical monuments and museums in Qatar during the 2022 FIFA
Birolini, S., Malighetti, P., Redondi, R., Deforza, P., 2019. Access mode choice to low-cost
World Cup. Transp. Policy 127 (August 2021), 116–129. https://doi.org/10.1016/j.
airports: evaluation of new direct rail services at Milan-Bergamo airport. Transp.
tranpol.2022.08.022.
Policy 73, 113–124. https://doi.org/10.1016/J.TRANPOL.2018.10.008.
Moeckel, R., Fussell, R., & Donnelly, R. (2014). Mode choice modeling for long-distance
Chen, T., & He, T. (2020). xgboost: eXtreme Gradient Boosting.
travel. Http://Dx.Doi.Org/10.1179/1942787514Y.0000000031, 7(1), 35–46.
Chen, X., Liu, Z., & Currie, G. (2016). Optimizing location and capacity of rail-based
https://doi.org/10.1179/1942787514Y.0000000031.
Park-and-Ride sites to increase public transport usage. Http://Dx.Doi.Org/10.1080/
Ortúzar, J. de D., & Willumsen, L. (2011). Modelling transport. https://books.google.
03081060.2016.1174366, 39(5), 507–526. https://doi.org/10.1080/
com/books?
03081060.2016.1174366.
hl=en&lr=&id=qWa5MyS4CiwC&oi=fnd&pg=PT7&ots=twYlYe0CSC&sig=-
Cherchye, L., Demuynck, T., De Rock, B., Hjertstrand, P., 2015. Revealed preference tests
UNBchao6uHrtrVUoabd8CraGUM.
for weak separability: An integer programming approach. J. Econ. 186 (1), 129–141.
Ouda, O.K.M., Shawesh, A., Al-Olabi, T., Younes, F., Al-Waked, R., 2013. Review of
https://doi.org/10.1016/J.JECONOM.2014.07.001.
domestic water conservation practices in Saudi Arabia. Appl. Water Sci. 3 (4),
Cirillo, C., Maness, M., & Serulle, N. U. (2014). Measuring value of travel time and travel
689–699. https://doi.org/10.1007/s13201-013-0106-1.
time variability in the presence of managed lanes: results from a pilot stated
preference survey on the Capital Beltway. Http://Dx.Doi.Org/10.1179/
15
Pasha, M. M., Hickman, M. D., & Prato, C. G. (2020a). Modeling Mode Choice of Air Model: Https://Doi.Org/10.1177/0361198118773556, 2672(47), 35–45. https://
Passengers’ Ground Access to Brisbane Airport: Https://Doi.Org/10.1177/ doi.org/10.1177/0361198118773556.
0361198120949534, 2674(11), 756–767. https://doi.org/10.1177/ Yaylali, M., Çelik, A. K., & Dilek, Ö. (2016). Analyzing key socio-economic and socio-
0361198120949534. demographic drivers of domestic passengers’ airline choice behavior in Turkey using
Petrik, O., Silva, J. de A. e, & Moura, F. (2016). Stated preference surveys in transport multinomial and mixed logit models. Http://Dx.Doi.Org/10.1179/
demand modeling: disengagement of respondents. Http://Dx.Doi.Org/10.1179/ 1942787515Y.0000000014, 8(3), 121–130. https://doi.org/10.1179/
1942787515Y.0000000003, 8(1), 13–25. https://doi.org/10.1179/ 1942787515Y.0000000014.
1942787515Y.0000000003. Yazdanpanah, M., Hosseinlou, M.H., 2016. The influence of personality traits on airport
Ramsey, E., Berglund, E.Z., Goyal, R., 2017. The impact of demographic factors, beliefs, public transport access mode choice: a hybrid latent class choice modeling approach.
and social influences on residentialwater consumption and implications for non- J. Air Transp. Manag. 55, 147–163. https://doi.org/10.1016/J.
price policies in urban India. Water (Switzerland) 9 (11), 844. https://doi.org/ JAIRTRAMAN.2016.04.010.
10.3390/w9110844. Zaidan, E., Abulibdeh, A., 2018. Modeling ground access mode choice behavior for
Sohoni, A.V., Thomas, M., Rao, K.V.K., 2017. Mode shift behavior of commuters due to Hamad International Airport in the 2022 FIFA World Cup city, Doha, Qatar. J. Air
the introduction of new rail transit mode. Transp. Res. Procedia 25, 2603–2618. Transp. Manag. 73, 32–45. https://doi.org/10.1016/j.jairtraman.2018.08.007.
https://doi.org/10.1016/J.TRPRO.2017.05.311. Zaidan, E., Abulibdeh, A., 2021. Master planning and the evolving urban model in the
Timothy, D. (2018). Routledge handbook on tourism in the Middle East and North Africa. gulf cities: principles, policies, and practices for the transition to sustainable
https://books.google.com/books? urbanism. Plan. Pract. Res. 36 (2), 193–215. https://doi.org/10.1080/
hl=en&lr=&id=XRypDwAAQBAJ&oi=fnd&pg=PT12&dq=Routledge+Handbook+ 02697459.2020.1829278.
on+Tourism+in+the+Middle+East+and+North+Africa&ots=wG- Zaidan, E., Abulibdeh, A., Alban, A., Jabbar, R., 2022. Motivation, preference,
wqScdKK&sig=ZSX5Ozh2x8DoKNLFq_uLOR4YEhY. socioeconomic, and building features: New paradigm of analyzing electricity
Tsamboulas, D.A., Nikoleris, A., 2008. Passengers’ willingness to pay for airport ground consumption in residential buildings. Build. Environ. 219, 109177. https://doi.org/
access time savings. Transp. Res. A Policy Pract. 42 (10), 1274–1282. https://doi. 10.1016/J.BUILDENV.2022.109177.
org/10.1016/J.TRA.2008.03.013. Zhu, P., Ho, S.N., Jiang, Y., Tan, X., 2020. Built environment, commuting behaviour and
Tseng, Y.Y., Knockaert, J., Verhoef, E.T., 2013. A revealed-preference study of job accessibility in a rail-based dense urban context. Transp. Res. Part D: Transp.
behavioural impacts of real-time traffic information. Transport. Res. Part C: Emerg. Environ. 87, 102438 https://doi.org/10.1016/J.TRD.2020.102438.
Technologies 30, 196–209. https://doi.org/10.1016/J.TRC.2011.11.006. Zopluoglu, C. (2019). Detecting examinees with item preknowledge in large-scale testing
Wang, F., & Ross, C. L. (2018). Machine Learning Travel Mode Choices: Comparing the using extreme gradient boosting (XGBoost): Https://Doi.Org/10.1177/
Performance of an Extreme Gradient Boosting Model with a Multinomial Logit 0013164419839439, 79(5), 931–961. https://doi.org/10.1177/
0013164419839439.
16

Analysis of Mode Choice Affects From The Introduction of Doha Metro Using Machine Learning and Statistical Analysis (2023)

Uploaded by

Copyright:

Available Formats

Analysis of Mode Choice Affects From The Introduction of Doha Metro Using Machine Learning and Statistical Analysis (2023)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Analysis of Mode Choice Affects From The Introduction of Doha Metro Using Machine Learning and Statistical Analysis (2023)

Uploaded by

Copyright:

Available Formats

Transportation Research Interdisciplinary Perspectives 20 (2023) 100852

Contents lists available at ScienceDirect

Transportation Research Interdisciplinary Perspectives

1. Introduction transportation, and as a solution to the problems of urban traffic

E-mail address: [email protected].

Fig. 2. Qatar Metro lines and stations.

Fig. 3. The research flowchart.

vector or simplified features, M is the maximum coalition size, ∅i ∊ ℝ is

Car 78.6% 1 16.3% Airport long-term car park 16.7% 1 34.5%

Number of persons traveling Education Nationality Number of persons in household

Male 53.4% Yes 4.2%

Training Error Testing Error Training Error Testing Error

Training Error Testing Error Training Error Testing Error

Accuracy Precision Recall F1-score Accuracy Precision Recall F1-score

Total 80.893% 81.97% 81.98% 81.77% 83.26% 83.27% 83.15%

Accuracy Precision Recall F1-score Accuracy Precision Recall F1-score

Total 80.32% 82.38% 80.32% 82.32% 87.54% 88.83% 87.53% 86.99%

Fig. 4. Interpretation of features of binary classification.

Fig. 5. Interpretation of features of multiclass classification.

Table 9 most important socioeconomic features and trip characteristic factors in

Number of travelers (5 travelers or more = reference category)

You might also like