Page 133 - Proceeding of Atrans Young Researcher's Forum 2019_Neat

“Transportation for A Better Life:
Smart Mobility for Now and Then”

23 August 2019, Bangkok, Thailand

choice analysis is traditionally used to analyze and alternatives and influenced factors. Those gaps
predict travel decision-making. It is a probabilistic cannot be solved using traditional techniques.
model to predict probabilities of different effects Methods from the field of Artificial Intelligence (AI)
with binomial model (with only two alternatives) or are a promising alternative to statistical approaches
multinomial model (with more than two for modeling travel decisions. Instead of making
alternatives). Therefore, discrete choice analysis is strict assumptions about the data, AI models learn to
utilized to predict the probability that an alternative represent complex relationships in a data-driven
is chosen, given certain values of input variables manner [13]. The usefulness of AI models has
[6,7] which collected through travel interview already been demonstrated for different areas in
surveys [8]. However, the accuracy of travel transport research. For example, AI models are
decision analysis using discrete choice requires particularly useful for classifying travel modes and
considerable efforts in conducting travel surveys. inferring trip purposes from global position system
For instance, Limtanakool et al. [9] employed data and acceleration data [14]–[16]. Other examples
from the Netherlands National Travel Survey to include the prediction of railway passenger demand
address the question as to how socioeconomic [17] and bimodal modeling of freight transport [18].
factors, land use attributes, and travel time affect Among many data-focused approaches, artificial
mode choice for medium- and longer- distance neural network (ANN), neuro-fuzzy or fuzzy time
travel, and how their role varies across trip purposes: series have been often employed in transport
commuting, business, and leisure. National Travel applications [19]–[21]. Mostafaeipour et al. [22]
Survey usually requires yearly collected data that predicted air travel demand in Iran by using a hybrid
cost and time consuming and necessitates the ANN with Bat and Firefly algorithms. Wang [20],
involvement of many different sectors. In another Yu and Schwartz [23] predicted tourism demand
research, Choudhury et al. [10] used a using fuzzy time series and hybrid grey theory. In a
comprehensive stated preference survey design and research of Sayed and Razavi [24], a neuro-fuzzy
a large choice set (with 9 modes, 5 departure times, approach was proposed for behavior mode choice
and 2 occupancy levels leading to 135 alternatives to modeling and confirmed that the neuro-fuzzy
investigate the acceptability of three new and approach was capable of producing the accuracy
emerging smart mobility options and quantify the obtained by the conventional methods and ANNs. In
associated willingness-to-pay values in the context short, AI has widely used to model a relationship
of Lisbon. Psaraki and Abacoumkin [11] estimated between the cause and effect of different real-life
the capacity requirements for the new Athens scenarios by combining the available data with
International Airport and forecasted the share of assumptions and probabilities for a better analysis.
each transport mode that airport passengers use by Nonetheless, AI methods are still underrepresented
conducting a passenger survey specifically designed in research of travel decision modeling.
for this purpose. Hess and Polak [12] investigated Hybrid AI methods namely the Ensemble
the effectiveness of many parking policy measures Decision Trees coupling with Bagging (EDT
through the analysis of parking choice behavior, Bagged) or Support Vector Machine (SVM) have
based on a stated preference dataset, collected in been widely used to solve various scientific problem.
various city center locations in the UK. Shortly, Over the years, the Decision Tree (DT) has been
countless efforts regarding time and budget have widely used in many applications such as face
been made to conducting travel interview surveys. detection [25], churn prediction and insurance fraud
Collecting a big volume of data set requires long detection [26] or diagnosis of heart disease [27].
planning process that might negatively influence on Having the same recognition, the SVM has been
transport demand modeling. Moreover, the quality utilized to solve many real world problems such as
of such surveys is highly influenced by the handwriting recognition [28], time series prediction
commitment of interviewers and the readiness of using least square support vector machine [29],
interviewees. intrusion detection [30], and prediction of membrane
Thusly, there is considerable number of studies protein types [31]. However, their application in
on discrete choice analysis to predict travel travel decision-making process have been largely
decisions, there are still gaps regarding the unexplored. A significant question whether such
uncertainties in conducting travel data surveys and methods are able to capture and incorporate the
modeling a non-linear correlation between travel

108