Comparative Analysis of Data Mining Classification Techniques for
Prediction of Problematic Internet Shopping
XUAN-LAM DUONG1, SHU-YI LIAW2*
1Department of Business Administration, Thai Nguyen University of Agriculture and Forestry
Quyet Thang Commune, Thai Nguyen City, Thai Nguyen Province
VIETNAM
2Department of Business Administration, National Pingtung University of Sciene and Technology
No. 1, Shuefu Road, Neipu, Pingtung 912301
R.O.C TAIWAN
Abstract: As online shopping has surged, so do disorders on internet purchasing. This study aims to develop and
compare predictive models that use data mining methods to predict problematic internet shopping. We used the
Artificial Neural Network (ANN), CHAID with bagging, and C5.0 and compared them with traditional logistic
regression to construct predictive models on a training cohort of 858 shoppers. Another cohort of 368 buyers was
utilized to confirm the accuracy of the predictive model. The accuracy, sensitivity, specificity, and the ROC-
AUC were used to assess the predictive performance. The C5.0 algorithm provided better accuracy in predicting
PIS than the other models, indicating that C5.0 might be a practical auxiliary algorithm for predicting PIS. Our
research findings cater to a comprehensive PIS prediction system, providing timely intervention and appropriate
support to individuals with the PIS problem.
Key-Words: problematic internet shopping, machine learning, data mining, online shopping
Received: February 15, 2024. Revised: April 14, 2024. Accepted: May 14, 2024. Published: June 14, 2024.
1 Introduction
Online shopping has become a frequently-used
alternative to traditional brick-and-mortar stores by
offering substantial convenience and benefits to
people's lives. However, for the minority population,
the buying-shopping urge becomes uncontrolled or
excessive, causing severe financial and psychological
consequences in individuals' routines [1]. Extant
literature on psychological and consumer behavior
has proposed several terms to characterize
problematic buying-shopping behavior, including
compulsive buying [2], shopping addiction [3, 4],
pathological buying [5], and buying-shopping
disorder [6]. Research has shown that specific
internet attributes such as availability, anonymity,
accessibility, and affordability contribute to
developing and maintaining an online subtype of
buying shopping disorder [5, 7].
Research has scrutinized problematic buying-
shopping from different prospects, and so far, no
agreed-upon definition has been reached. This broad
etiological spectrum adds more complexity to the
theoretical explanation and requires further empirical
evidence to determine global diagnostic criteria for
problematic buying-shopping. Although problematic
internet shopping has not been formally included in
any monopoly classification of diseases, it has been
hypothesized as a behavioral addiction in the
literature [1, 3]. Pathological buying online – another
derivative of problematic online shopping has been
postulated as a sub-type of Internet addiction [5] that
might adversely influence one's daily and social
routine and economic status [3]. The authors are on
the side, supporting that a more neutral term that does
not directly imply that the behavior is addictive
would be better when referring to uncontrolled and
excessive online behavior. Therefore, we use
"problematic internet shopping," a more neutral
expression in agreement with previous studies [8, 9],
to refer to the online version of problematic buying-
shopping behavior.
The growing incidence of problematic internet
buying/shopping requires a quick and efficient
prediction system [10]. However, thus far, no study
has employed data mining algorithms and techniques
to detect unregulated online buying/shopping
behavior. Therefore, the current study sought to
develop predictive models and compare the
International Journal of Applied Sciences & Development
DOI: 10.37394/232029.2024.3.7
Xuan-Lam Duong, Shu-Yi Liaw
E-ISSN: 2945-0454
82
Volume 3, 2024
predictive performance of several classification
algorithms to provide a basis for early intervention
and proper support for those who might be at risk of
problematic internet buying/shopping.
2 Literature review
Big data has significantly influenced the electronic
commerce industry and will likely continue act in this
way. E-retailers are counting on cloud-based big data
analytics to harness the power of big data. Prior
research has applied machine learning algorithms and
data mining techniques to analyze problematic online
behaviors. The methods for predicting whether a
person can suffer from problematic online behavior
can benefit the medical field and individuals.
Nevertheless, a handful of papers have tackled the
issue, such as Arora et al. [11], who discuss the role
of machine learning in assessing the addictive use of
various online technologies and its influence on
mental and emotional health.
According to a recent systematic review, a bulk of
studies in addiction research employed machine
learning to predict substance addiction [12], leaving
a handful of research that has tackled the issue of
internet-related addictive behaviors, such as internet
addiction [13, 14] or problematic smartphone use
severity [15]. Efforts have been made to implement
machine learning in diagnosing and detecting
problematic buying or shopping using different
methods or combining it with several algorithms to
enhance accuracy [16]. For example, Prashar et al.
[17] employ multiple machine learning classifiers to
predict impulsive buying behavior. Their findings
suggest the superiority of logistic regression
regarding predictive power to other techniques. In
contrast, Prashar and Mitra [18] provide statistical
evidence that SVM surpasses logistic regression,
linear discriminant analysis, quadratic discriminant
analysis, and k-Nearest Neighbor methods in
predicting power. Problematic internet buying or
shopping is becoming prevalent in our consumer
society. Surprisingly, less is known about the
predictivity of problematic internet shopping using
data mining algorithms.
3 Methodology
3.1 Data collection and sample
An online questionnaire was designed and
administered to acquire demographic information,
internet usage habits, and perceived online shopping
benefits and risks. The authors developed an online
questionnaire that might take approximately 20
minutes to comprehend. The first part of the
questionnaire, aiming at collecting general
demographic characteristics and everyday internet
use statistics, is followed by a battery of validated
scales (see Table 1). After pre-testing to mitigate any
ambiguous wording or administering problems, the
revised questionnaire was disseminated online to
internet users, those who had shopped online over
their last twelve months via communication
applications and social networks, such as Facebook,
Line, and WeChat. The cover of the questionnaire
contains information about the purposes of the study.
Respondents are asked to provide written informed
consent before proceeding to the body of the
questionnaire. Data were collected anonymously and
treated confidently. The author put numerous efforts
to improve the quality of the data. The online survey
management system enables us to identify duplicate
responses and records that contain unusual patterns
(i.e., straight-lining or answers completed in an
abnormally instant manner), enhancing the accuracy
and appropriateness of the data. There is no missing
data in our dataset. Finally, the data screening and
outliers removal procedure resulted in 1,226 eligible
respondents available in our dataset.
3.2 Measures
We adopt validated scales measuring the perceived
benefits and risks of online shopping. All measures
were assessed on a seven-point Likert-type scale
where 1 = strongly disagree, 7 = strongly agree.
Cronbach's alpha and McDonald's omega are used to
evaluate the scale's reliability, whereas the latter is
used to overcome the deficiencies of alpha. The
analysis results showed that the Cronbach's 𝛼 of all
variables ranged from 0.759 to 0.897 while the
McDonald's’ 𝜔 ranged from 0.770 to 0.898. The
minimal differences between the two measures
demonstrate that the scales have adequate reliability
[19].
The Online Shopping Addiction Scale [22, 23] was
adapted to measure problematic internet shopping
severity. The 18-item Likert-type scale measures
problematic internet shopping based on six core
components of addictive behaviors – i.e., salience,
mood modification, tolerance, withdrawal
symptoms, conflict, and relapse [24]. The mean
scores and their corresponding standard deviations
were used to group respondents into two categories -
i.e., regular and problematic internet shoppers using
the cut-off score of one standard deviation above the
mean.
International Journal of Applied Sciences & Development
DOI: 10.37394/232029.2024.3.7
Xuan-Lam Duong, Shu-Yi Liaw
E-ISSN: 2945-0454
83
Volume 3, 2024
Afterward, the authors submitted eight variables
measuring the benefits and risks of online shopping
(Table 1) and respondents' demographic
characteristics and treated them as input variables in
the predictive models. This information includes age,
gender, marital status, education level, internet
experience, daily internet usage, daily internet
shopping usage, and monthly budget for internet
shopping. All the continuous variables were
standardized to enhance the interpretation capability
(e.g., age, internet experience, internet usage, and
internet shopping usage). Categorical variables were
transformed into numerical values. Data mining
prediction models were constructed using SPSS
Modeler version 18.0 (SPSS Inc., Chicago, IL, USA).
3.3 Data analysis and model building
The authors employed three supervised machine
learning algorithms, namely Artificial Neural
Network (MLP) with bagging, CHAID, and C5.0, to
construct the predictive models using the holdout
testing method. The performance was then used to
compare with the traditional Logistic regression. The
original dataset was arbitrarily split into two sets,
with the training dataset comprising about 70% of the
respondents (n = 858) and the testing dataset
including 30% of the participants (n = 368). The PIS
score was treated as a target variable, whereas the
sum score of eight factors of internet purchasing,
demographic characteristics and internet use patterns
were treated as input variables in predictive models.
We consult the overall accuracy and additional
metrics such as ROC curves, AUC, and Gini to assess
the predictive performance.
4 Results
4.1 Descriptive statistic
The sample comprises 519 (42.33%) males and 707
females (57.67%). The mean age was 31.28
(SD=9.81), with a median of 29, ranging from 17 to
70. Regarding marital status, 587 (47.88%) were
single or without stable partners; 639 (52.12%)
reported being in a relationship. From the educational
level point of view, 403 (32.87%) attained high
school diplomas, 627 (51.14%) obtained a
university/college degree, and 196 (15.98%) owned a
graduate’s degree. On average, respondents have
more than 13 years of internet experience. They spent
over 6.6 hours on internet use but only consumed
approximately 1.7 hours per day for online shopping-
related activities. More than 85% of the respondents
reported spending less than $1,000 for online
shopping every month. This sample is deemed
appropriate because younger people, females, and
those with higher online shopping frequency are
more prone to manifest problematic internet
shopping.
4.2 Predictive performance
Table 2 shows the results from the predictive models.
ANN C5.0, and CHAID outperform logistic
regression in all performance evaluation statistics in
the training dataset.
The Artificial Neuron Network (MLP) achieved a
classification accuracy of 77.51% with a sensitivity
of 80.46% and a specificity of 74.46%; the CHAID
with bagging achieved a classification accuracy of
84.73%, with a sensitivity of 84.37% and a
specificity of 85.11%. However, the C5.0 classifier
performed best among the four evaluated models.
The C5.0 model had a classification accuracy of
86.06%, with a sensitivity of 85.16 and a specificity
of 88.09%. The AUC values across four models
ranged from 0.769 (LR) to 0.942 (C5.0), and the Gini
ranged from 0.537 to 0.885, respectively.
Similar patterns can be observed in the testing
dataset, where the decision tree (C5.0) outperforms
other models in all respective performance
indicators.
Table 1. List of measures
Variables
Number of
items
Cronbach's 𝛼
McDonald's 𝜔
References
Information search
4
0.852
0.854
[20]
Recommendation system
4
0.816
0.821
[20]
Dynamic pricing
4
0.782
0.786
[20]
Customer service
4
0.883
0.883
[20]
Privacy
4
0.767
0.770
[20]
Security
4
0.759
0.765
[20]
Group influence
4
0.897
0.898
[20]
Deception
4
0.866
0.867
[21]
International Journal of Applied Sciences & Development
DOI: 10.37394/232029.2024.3.7
Xuan-Lam Duong, Shu-Yi Liaw
E-ISSN: 2945-0454
84
Volume 3, 2024
The CHAID model appeared to be the worst
prediction model, with a classification accuracy of
66.30%, lower than the 68.75% obtained from
Logistic regression. Also, there are minimal
differences between the sensitivity and specificity of
the CHAID and the logistic regression models. The
C5.0 predictive decision tree achieved the highest
accuracy of 82.07%, with 82.22% sensitivity and
81.91% specificity. The AUC across models ranges
from 0.731 to 0.921, while the Gini ranges from
0.462 to 0.843. In the whole sample, the accuracy,
sensitivity, and specificity of the logistic regression
model were 69.33%, 70.21%, and 68.43%,
respectively. While predictive performance
differences between the Neural Network and the
CHAID are minimal yet, the C5.0 achieved the
highest accuracy (85.24%), sensitivity (84.38%), and
specificity (86.11%). Also, the AUC and Gini of C5.0
were superior comparing to the others.
As shown in Figure 1 for the training sample, the four
predictive models manifested differently, except for
the substantial convergence between CHAID and the
C5.0. Also, the CHAID and the C5.0 achieved the
highest predictive performance, whereas the logistic
regression showed the worst predictive capability.
Figure 1. ROC curve of four classifiers
While the C5.0 maintains its strong forecasting
capability for the testing sample, the other three
models exhibit a minimal distinction.
4.3 Predictor importance
The sensitivity analysis was performed where
each variable is placed in order of its relative
importance, giving the objective is to determine the
relative importance of each of the 16 independent
variables within different models. The results from
the sensitivity analysis are presented in Table 3.
Table 2. The performance of predictive models
Dataset
Logistic regression
Artificial Neural
Network (MLP)
CHAID
Decision tree
(C5.0)
C+
C-
C+
C-
C+
C-
C+
C-
Training dataset
True Positive
308
127
350
85
367
68
373
65
True Negative
134
289
108
315
63
360
50
370
Accuracy (%)
69.58
77.51
84.73
86.60
Sensitivity (%)
70.80
80.46
84.37
85.16
Specificity (%)
68.32
74.46
85.11
88.09
AUC
0.769
0.849
0.925
0.942
Gini
0.537
0.699
0.851
0.885
Testing dataset
True Positive
128
58
141
45
124
62
148
32
True Negative
57
125
66
116
62
120
34
154
Accuracy (%)
68.75
69.84
66.30
82.07
Sensitivity (%)
68.82
75.81
66.67
82.22
Specificity (%)
68.68
63.74
65.93
81.91
AUC
0.731
0.749
0.740
0.921
Gini
0.462
0.498
0.480
0.843
Total dataset
True Positive
436
185
491
130
491
130
524
97
True Negative
191
414
174
431
125
480
84
521
Accuracy (%)
69.33
75.20
79.20
85.24
Sensitivity (%)
70.21
79.06
79.06
84.38
Specificity (%)
68.43
71.24
79.34
86.11
AUC
0.758
0.819
0.873
0.936
Gini
0.515
0.638
0.746
0.873
Note: C+ denotes the count of predictive positive; C- denotes the count of predictive negative
International Journal of Applied Sciences & Development
DOI: 10.37394/232029.2024.3.7
Xuan-Lam Duong, Shu-Yi Liaw
E-ISSN: 2945-0454
85
Volume 3, 2024
Table 3. The importance of the input variables in four
models
Ord
era
Logistic
regression
Neural
network
(MLP)
C5.0
CHAID
(bagging)
1
Shopping
usage
Shopping
usage
Shoppi
ng
usage
Security
2
Internet
experience
Security
Educati
on
Shopping
usage
3
Recommen
dation
system
Recommen
dation
system
Group
influenc
e
Recommen
dation
system
4
Security
Internet
experience
Age
Dynamic
pricing
5
Privacy
Information
search
Informa
tion
search
Information
search
6
Monthly
budget
Customer
service
Securit
y
Privacy
7
Deception
Dynamic
pricing
Decepti
on
Customer
service
8
Dynamic
pricing
Privacy
Privacy
Group
influence
9
Education
Age
Custom
er
services
Internet
experience
10
Information
search
Group
influence
Monthl
y
budget
Deception
11
Gender
Internet
usage
Marital
status
Monthly
budget
12
Internet
usage
Education
Gender
Gender
13
Marital
status
Deception
Dynami
c
pricing
Internet
usage
14
Age
Monthly
budget
Internet
usage
Education
15
Group
influence
Marital
status
RS
Marital
status
16
Customer
service
Gender
Internet
experie
nce
Age
aThe order according to importance, from the most to the
least important.
Accordingly, the predictor performance indicates that
daily online shopping time, security, and
recommendation systems were the most critical
predictors explaining PIS in our study samples. In
contrast, gender, age, marital status, and daily
internet usage manifested a modest role in predicting
problematic internet shopping. The excessive time
spent on internet shopping appears to be the most
critical indication of problematic internet shopping,
followed by the internet experience and the effects of
the recommender systems, which could influence
consumers' online shopping. Conversely, gender,
age, and marital status show a minimal explanation
for detecting individuals who might be at risk of PIS.
Prior studies have employed predictive models for
different purposes of interest. The results of this
study indicate that the C5.0 decision tree is the best
classifier, with 85.24% accuracy on the whole
dataset. The CHAID model came out second, with
79.20% accuracy, leaving the ANN model the
poorest performance among the three models with
75.20% accuracy. Overall, the predictive capability
of C5.0, CHAID, and ANN are much higher than the
logistic regression. These findings align with prior
investigations [25].
5 Conclusion
Our study constructed and compared three data
mining algorithms to predict the problematic internet
shopping behavior and found that decision trees –
i.e., C5.0 and CHAID outperformed Neural Network
and Logistic Regression in classifying consumers in
the 'at risk' group. Predicting those at risk of PIS is
crucial in management decision-making since timely
diagnosis is coupled with more advantageous
treatment outcomes. We expect data mining methods
such as C5.0 and CHAID could serve as effective
alternatives to conventional logistic regression in
identifying the critical variables more accurately and
timely. Nevertheless, it is worth scrutinizing more
complex machine learning algorithms (i.e., Random
Forest, Xgboost, etc.), albeit the predictive
performance of C5.0 and CHAID in the current study
prevails over other machine learning algorithms.
There are several limitations to our study. On the
one hand, the sample was collected from a
comparatively sizeable non-clinical sample with a
minimal possibility of obtaining information from
those diagnosed with PIS. The authors contemplate
that the strength of employing data mining
algorithms to specify the diagnostic criteria of PIS
may be more fully verified in more extensive or even
more diverse populations. A further limitation was
that the present study employs a scale measuring
problematic internet shopping from an addiction
perspective, a more adverse and stringent condition
that might result in fewer problematic internet
shoppers being detected than it would have been
capable of. We believe that a new scale that explicitly
measures problematic buying/shopping behavior
might be beneficial in detecting relevant
observations.
International Journal of Applied Sciences & Development
DOI: 10.37394/232029.2024.3.7
Xuan-Lam Duong, Shu-Yi Liaw
E-ISSN: 2945-0454
86
Volume 3, 2024
References:
[1] Lejoyeux, M., and Weinstein, A, “Shopping
Addiction,” Principles of Addiction, 847–
853. (2013).
[2] Weinstein, A., Maraz A, M., Griffiths, M. D.,
Lejoyeux, M., and Demetrovics, Z, “Compulsive
buying – features and characteristics of
addiction. In: Neuropathology of Drug
Addictions and Substance Misuse, 3, pp.993-
1007. Academic Press, 2016.
[3] Andreassen, C.S., Griffiths, M. D., Pallesen, S.,
Bilder, R. M., Torsheim, T., and Aboujaoude, E,
“The Bergen Shopping Addiction Scale:
reliability and validity of a brief screening test,”
Frontiers in Psychology, 6:1374 (2015).
[4] Rose, S., and Dhandayudham, A, “Towards an
understanding of Internet-based problem
shopping behavior: The concept of online
shopping addiction and its proposed predictors,”
Journal of Behavioral Addictions, 3(2), 83-89
(2014).
[5] Trotzke, P., Starcke, K., Müller, A., and Brand,
M, “Pathological buying online as a specific
form of Internet addiction: a model-based
experimental investigation,” PloS ONE, 10(10:
e0140296 (2015).
[6] Müller, A., Steins-Loeber, S., Trotzke, P., Vogel,
B., Georgiadou, E., and de Zwaan, M, “Online
shopping in treatment-seeking patients with
buying-shopping disorder,” Comprehensive
Psychiatry, 94: 152120 (2019).
[7] Dittmar, H., Long, K, and Bond, R, “Association
between materialistic values, emotional and
identity-related buying motives, and compulsive
buying tendency online,” Journal of Social and
Clinical Psychology, 26(3), 334-361 (2007).
[8] Ko, Y. M., Roh, S. W., and Lee, T. K, “The
association of problematic internet shopping with
dissociation among South Korean internet
users,” International Journal of Environmental
Research and Public Health, 17: 3235 (2020).
[9] Lam, L., and Lam, M. K, “The association
between financial literacy and Problematic
Internet Shopping in a multinational sample,”
Addictive Behaviors Reports, 6, 123-127 (2017).
[10] Gori, A., Topino, E., and Casale, S,
“Assessment of online compulsive buying:
Psychometric properties of the Italian
compulsive online shopping scale (COSS)”,
Addictive Behaviors, 129, 107274 (2022)
[11] Arora, A., Chakraborty, P., and Bhatia, M. P.
S, “Problematic Use of Digital Technologies and
Its Impact on Mental Health During COVID-19
Pandemic: Assessment Using Machine
Learning,” in I. Arpaci, M. Al-Emran, M. A. Al-
Sharafi, and G. Marques (Eds.), Emerging
Technologies During the Era of COVID-19
Pandemic, Springer International Publishing,
197-221 (2021).
[12] Mak, K.K., Lee, K., and Park, C,
“Applications of machine learning in addiction
studies: a systematic review,” Psychiatry
Research, 275, 53-60 (2019).
[13] Di, Z.L., Gong, X.L., Shi, J.Y., Ahmed,
H.O.A., and Nandi, A.K, “Internet addiction
disorder detection of Chinese college students
using several personality questionnaire data and
support vector machine,” Addictive Behaviors
Reports, 10: 100200 (2019).
[14] Ioannidis, K., Chamberlain, S.R., Treder,
M.S., Kiraly, F., Leppink, E.W., Redden, S.A.,
Stein, D.J., Lochner, C., and Grant, J.E,
“Problematic internet use (PIU): Associations
with the impulsive-compulsive spectrum. An
application of machine learning in psychiatry,”
Journal of Psychiatry Research, 83, 94-102
(2016).
[15] Elhai, J.D., Yang, H.B., Rozgonjuk, D., and
Montag, C, “Using machine learning to model
problematic smartphone use severity: The
significant role of fear of missing out,” Addictive
Behaviors, 103: 106261 (2020).
[16] Nawodya, A. G. S., and Kumara, B. T. G. S,
“Machine learning approach to detect online
shopping addiction, 2nd International Conference
on Advanced Research in Computing (ICARC),
Belihuloya, Sri Lanka (2022).
[17] Prashar, S., Parsad, C., and Vijay, T. S,
“Predicting impulsive buyers: A comparative
study of binary classifiers' discriminative
ability,” International Journal of Strategic
Decision Sciences (IJSDS), 7(2), 1-17 (2016).
[18] Prashar, S., and Mitra, S. K, “Forecasting
impulsive buying behavior: a comparative study
of select five statistical methods,” International
Journal of Business Forecasting and Marketing
Intelligence, 3(3), 289-308 (2017).
[19] Hair, J, F., Babin, B, J., Anderson, R. E., and
Black, W. W, Multivariate Data Analysis 8th
Edition. CENGAGE (2019).
[20] Le, T.M and Liaw, S.-Y, “Effects of Pros and
Cons of Applying Big Data Analytics to
Consumers’ Responses in an E-Commerce
Context,” Sustainability, 9, 798 (2017).
[21] Román, S, “The ethics of online retailing: A
scale development and validation from the
consumers’ perspective,” Journal of Business
Ethics, 72, 131-148 (2007).
[22] Zhao, H-Y., Tian, W., and Xin, T, “The
development and validation of the Online
International Journal of Applied Sciences & Development
DOI: 10.37394/232029.2024.3.7
Xuan-Lam Duong, Shu-Yi Liaw
E-ISSN: 2945-0454
87
Volume 3, 2024
Shopping Addiction Scale,” Frontiers in
Psychology, 8:375 (2017).
[23] Duong, X-L., and Liaw, S-Y, “Psychometric
evaluation of Online Shopping Addiction Scale
(OSAS)”, Journal of Human Behavior in the
Social Environment, 32 (5), 618-628 (2023)
[24] Griffiths, M, “A ‘components’ model of
addiction within a biopsychosocial framework,”
Journal of Substance Use, 10(4), 191-197
(2005).
[25] Meng, X-H., Huang, Y-X., Rao, D-P.,
Zhang, Q., and Liu, Q, “Comparison of three data
mining models for predicting diabetes or
prediabetes by risk factors,” Kaohsiung Journal
of Medical Sciences, 29, 93-99 (2013).
Contribution of Individual Authors to the
Creation of a Scientific Article (Ghostwriting
Policy)
The authors equally contributed in the present
research, at all stages from the formulation of the
problem to the final findings and solution.
Sources of Funding for Research Presented in a
Scientific Article or Scientific Article Itself
No funding was received for conducting this study.
Conflict of Interest
The authors have no conflicts of interest to declare
that are relevant to the content of this article.
Creative Commons Attribution License 4.0
(Attribution 4.0 International, CC BY 4.0)
This article is published under the terms of the
Creative Commons Attribution License 4.0
https://creativecommons.org/licenses/by/4.0/deed.en
_US
International Journal of Applied Sciences & Development
DOI: 10.37394/232029.2024.3.7
Xuan-Lam Duong, Shu-Yi Liaw
E-ISSN: 2945-0454
88
Volume 3, 2024
