Page 137 - 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

shops, buying goods very quickly then moving  Original dataset is randomly divided into a
away. Additionally, the real parking charge was also training dataset (in this case 70% of the initial
zero in many cases. Most of commuters had free dataset) and a testing dataset (30% of the initial
parking since they were paid by their companies dataset).
while the parking charge of shoppers was included  Construction of a classification tree is done
in the receipts as the bonus for their shopping. from the training dataset using cross-validation.
 A bootstrap replicate is then taken from the
2.3. Methods used initial dataset to generate the new subset, a new
In this section, the algorithm of both EDT version of classifier is then trained using this
Bagged as well as SVM were introduced. subset. This step is repeated as many times as
the number of subtrees reach the certain number
2.3.1. Ensemble Decision Trees – Bagged Trees of trees which we want to create.
 The bootstrap aggregation is then calculated by
Ensemble Decision Trees – Bagged Trees (EDT averaging the results of all trained classifiers.
Bagged) methods are generally composed of several This step is done using the following equation;
decision trees [49] in order to perform better
N
prediction results than a single decision tree. In EDT f   x  1  f *   x (1)
Bagged, Bagging is a technique used to assemble bag N i 1 i
several versions of a decision tree to create an
ensemble of classifiers [50]. The multiple version of where x is the input variables and N is the number of
a decision tree is created by replicating bootstraps subtrees constructed during the algorithm.
from the initial dataset and use them as new learning To construct the EDT Bagged for predicting
datasets. Each subset of the original data set is then travel decisions of transport users, in this study, a k-
used to train the model and we obtain at the end an fold cross-validation was applied to assess the
ensemble of trained models. These models then performance of the model with the number of folds
combine to give the final result. The EDT Bagged was chosen as 15 folds and the number of learning
algorithm consists of 4 main steps as the followings cycles was chosen as 30.
(Figure 1):

2.3.2. Supported Vector Machine
Figure 1. EDT Bagged algorithm for Support Vector Machines (SVM) is a machine
classification problem learning algorithm based on statistical learning
theory. It was first introduced by Vapnik et al. [51],
[52]. In this paper, we used SVM as a classification
method in order to predict the travel decisions of

112