IAES International Journal of Artificial Intelligence (IJ-AI)

Vol. 13, No. 4, December 2024, pp. 4273~4280

ISSN: 2252-8938, DOI: 10.11591/ijai.v13.i4.pp4273-4280  4273

Elevating fraud detection: machine learning models with

computational intelligence optimization

Cheryl Angelica, Charleen, Antoni Wibowo

Department of Computer Science, Master of Computer Science, Bina Nusantara University, Jakarta, Indonesia

Article Info ABSTRACT

Article history: The amount of crimes committed online has undoubtedly increased as more
people use the internet for e-commerce and other financial transactions.
Received Jan 4, 2024 Machine learning algorithms have been created to detect payment fraud in
Revised Mar 26, 2024 online purchasing in order to address the issue. This study performs a
Accepted Apr 12, 2024 thorough comparative examination of different metaheuristic optimizations
as hyperparameter tuning methods; these are particle swarm optimization
(PSO) and genetic algorithm (GA). They are used to optimize the receiver
Keywords: operating characteristic (ROC) area under the curve (AUC) of the three
machine learning algorithms, namely X-gradient boost, random forest
E-commerce classifier, and light gradient boost machine. Since the study's data are
Fraud detection unbalanced, the determined metrics were ROC AUC. PSO offers consistent
Genetic algorithm conditions for finding the best solution, according to our experiment.
Light gradient boost machine Without the inclusion of population annihilation strategies, PSO can achieve
Particle swarm optimization the greatest results in various situations which are different from GA, a
Random forest classifier consistent condition for finding the best solution, according to our
X-gradient boost experiment. Without the inclusion of population annihilation strategies, PSO
can achieve the greatest results in various situations. The findings indicate
that random forest classifier provided the highest ROC AUC value both
before and after the hyperparameter tuning process, with a score of 88.69%
attained while utilizing PSO.
This is an open access article under the CC BY-SA license.

Corresponding Author:
Charleen
Computer Science Department, Master of Computer Science, Bina Nusantara University
Jakarta 11480, Indonesia
Email: [email protected]

1. INTRODUCTION
The way society functions has undergone a significant change as a result of technological
advancement. The development of technologies has completely changed how people could go about their
everyday life and do business. Meeting personal needs is one of the improvements. They only purchase their
wants using digital technologies namely online shopping. It has become a popular and practical way to buy
goods and services. The number of options for online shopping has significantly increased with the
development of companies like Shopee and Tokopedia. However, the increasing usage of the internet for
e-commerce and other financial operations has unavoidably increased the number of crimes committed on the
internet. The concealment of the complex network may provide a breeding ground for criminals to engage in
fraudulent activities [1].
The merchants must deal with a wide variety of fraud and abuse. The account takeover, free-trial
abuse, refund fraud, reseller fraud, fake product news, first-party fraud, payment fraud, and many more are
examples of fraud [2]. All of these put the merchants' reputation and revenue at risk. Money laundering and
the spread of fake news are two examples of fraud that have serious repercussions for society as a whole.

Journal homepage: http://ijai.iaescore.com

4274  ISSN: 2252-8938

E-commerce fraud is increasing quickly because, unlike in the early years of the internet, today's scammers
are well-funded and well-equipped professional rings. The extremely constrained window of time in which
acceptance or denial must be made is the main difficulty in fraud detection. The sheer volume of transactions
that must be handled at one time is also remarkable. One of the main concerns with e-commerce systems is
identifying fraud as soon as possible when the transaction is being performed [3]. In reality, fraudulent and
legitimate transactions coexist, which makes it challenging to accurately detect them using basic pattern
matching methods [4]. Therefore, this essay focuses on how to spot payment fraud in online shopping.
For the purpose of identifying fraud, a lot of machine learning models have been developed and
have been shown to be capable of saving billions of dollars annually. Machine learning is equivalent to
having multiple groups of analysts run thousands of queries and evaluate the outcomes to choose the optimal
solution. When identifying minor trends or illogical patterns, it is much faster, more scalable, effective, and
more accurate than humans. This indicates that the model is able to recognize suspect clients even in the
absence of a chargeback.
Hyperparameter tuning is done to improve the machine learning algorithm. Metaheuristic
optimization techniques, including particle swarm optimization (PSO) and genetic algorithm (GA), were
employed in this research. The basic concept of GA's algorithm is the theory of life evolution. PSO, on the
other hand, adopts the theory of colonization from a swarm of feeding creatures [5]. The layout of this essay
will be as follows. Related works to this study are included in section 2. The experiment's technique or
methodology was explored in section 3. While part 4 will focus on the experiment results analysis. Finally,
the conclusion was covered in section 5.
In machine learning, there are various methods for detecting fraud. The effectiveness of fraud
detection software is frequently correlated with the efficiency of utilized feature engineering strategies,
particularly those that explain client behavior because this information is quite helpful in identifying fraud
patterns. Fraud by credit cards is rising at the same rate as the number of customers.
Khatri et al. [6] tested the suitability of supervised machine learning models for estimating the
likelihood of fraudulent transactions using an unbalanced dataset. They studied some models including
random forest, K-nearest neighbor (KNN), decision tree, naive Bayes, and logistic regression. The decision
tree model is the most effective one for foretelling these frauds. Although the KNN model is more sensitive
than the decision tree, it takes much longer to test the data. Another approach is using machine learning
algorithms like the random forest and the Adaboost algorithms. Sailusha et al. [7] classify the credit card
transactions that are within the dataset as both fraud and non-fraud transactions. In terms of recall, precision,
and F1-score, the random forest algorithm surpasses the Adaboost algorithm.
Deep learning has become an important factor among several subfields of finance. Nanduri et al. [2],
suggested a novel deep learning-based model for rules extraction to identify fraudulent e-commerce
transactions. This algorithm surpasses the state-of-the-art in terms of recall, accuracy, area under the receiver
operating characteristic (AUROC), and F1-score. The absence of all likely patterns needed to train adequate
supervised learning models is commonly the difficulty related with fraud detection. The fact that fraudulent
patterns are not just rare but also change over time makes this problem worse. Youssef et al. [8] propose a
method that, by giving each data point a consistency score, has a great deal of promise for detecting outliers
and pure inliers.
Taobao, recognized as one of the largest global e-commerce platforms, boasts a substantial repository
of transaction data. The platform employs a sophisticated fraud detection system, extensively evaluated in [9].
The assessment reveals that the proposed attacks exhibit a notable likelihood of circumventing the deployed
detector, causing a significant reduction in average precision from nearly 90% to 20%.
Microsoft’s new fraud-management system (FMS) explained in [10] can quickly identify and stop
new fraud patterns. Advanced real-time dynamic risk feature generation methods and specially created
short- and long-term sequential machine learning models are used. The system also optimizes the risk-control
module's decisions using dynamic optimization techniques known as prospective control modeling for the
highest long-term sales and profits. An easy way to comply with the conference paper formatting
requirements is to use this document as a template and simply type your text into it.

2. PROPOSED METHODS
The phases of this research are as follows: i) a preliminary study; ii) data gathering; iii) data pre-
processing; and iv) construction of the machine learning models and evaluation. A preliminary study is made
in order to fully comprehend the subject of e-commerce fraud, which has been chosen for the research. In the
second stage and third stage, the data are gathered and pre-processed to make them appropriate for machine
learning models. The various machine learning models will next be compared with our suggested model
which is the model with the addition of a meta-heuristic algorithm as their hyperparameter tuning methods.

Int J Artif Intell, Vol. 13, No. 4, December 2024: 4273-4280

Int J Artif Intell ISSN: 2252-8938  4275

2.1. Preliminary study

This research's initial phase was a preliminary investigation that included literature reviews. The
purpose of the literature review was to identify prior research, clarify the theoretical underpinnings, and
describe the methodologies employed in this study. Reading sources from journals relating to machine
learning models and e-commerce fraud detection, like IEEE and Springer, are searched and reviewed as part
of the study of literature.

2.2. Data gathering

The IEEE-computational intelligence society (IEEE-CIS) provided the dataset for this study, which
was made accessible by Vesta Corporation. The pioneer in secure online payment methods is Vesta
Corporation. Identity table and transaction table are two of the data sets that are utilized to create the machine
learning model in this study. A transaction table has 590,540 records and 394 feature columns. The
transaction table includes information about money transfers as well as other gifting items and services,
including product codes, payment card details, client addresses, and transaction amounts. 41 columns of
features make up the 144,233 data in the identity database. A digital signature (UA/browser/os/version) and
network connection information (IP, ISP, proxy) related to transactions are both stored in the Identity table.
They are gathered by the partners in digital security and Vesta's fraud prevention system.

2.3. Data pre-processing

Data pre-processing comes next after data collection. Raw data must be modified during data
pre-processing in order for it to be used in a machine learning model. It involves data cleaning, data
formatting, and data transformation. Dropping the columns with more than 20% of the values missing is the
first step in cleaning and formatting the data in this dataset. Then, the category and numerical features are
divided and handled differently depending on the kind of each feature. The approach to filling in the missing
values for numerical attributes is to use the median. For categorical features, however, the most common
value should be used to fill in any missing values. Then, the category and numerical properties are combined
once more. For data transformation, by comparing each category level to a predetermined reference level,
one-hot encoding is used to convert the categorical features for data transformation [11]. The dataset was
then divided into a 20% testing set and an 80% training set. While StandardScaler function is used to
standardize the functionality of the input dataset for numerical features.
The dataset has an imbalance class for classification, as shown by the data. Figure 1 representation
of the change of dataset's high normal-to-fraudulent ratio illustrates how predictions from the original
training set were wildly distorted to balanced dataset. Imbalanced class classification of data is a significant
issue for machine learning and hard to improve accuracy. Almost every classification technique will provide
substantially higher accuracy for the majority class than for the minority class when working with
unbalanced data. An indication of subpar categorization performance is this disparity. In order to resolve this
issue, the data was processed with the synthetic minority over-sampling technique (SMOTE) algorithm by
oversampling method. Modifying the majority or minority class distribution allows the classification
algorithm to treat an imbalanced dataset equally [12]. By resampling the minority class sample, this strategy
creates a fresh sample from the minority class to balance the dataset [13].

Figure 1. Pie chart of total fraud and not fraud before SMOTE and after SMOTE

Elevating fraud detection: machine learning models with computational intelligence … (Cheryl Angelica)
4276  ISSN: 2252-8938

2.4. Model building and evaluation

Model building and evaluation come next. Figure 2 represents the flow of creating the model. Our
approach to determine the optimal machine learning model for spotting ecommerce fraud involves comparing
models like random forest, light gradient boost machine, and x-gradient boost. To improve the model's
accuracy and receiver operating characteristic (ROC) area under the curve (AUC) value, hyperparameter
approaches will be integrated with it. Two hyperparameter optimization approaches, PSO and GA, were
applied to enhance the model's quality. John Holland invented GAs, which were influenced by Darwin's
theory of evolution. A search algorithm based on the principles of natural selection and genetics is known as
a GA [14]. PSO, a stochastic optimization technique, is based on the movement and behavior of swarms.
PSO uses the concept of social engagement to resolve a conflict. It makes use of a swarm of particles (called
agents) that roam the search space in search of the optimum solution. The positional coordinates in the
solution space that correspond to each swarm member's best solution thus far are sought after by each
member of the swarm. It is known as a personal best [15]. The optimum hyperparameter approaches for each
machine learning model will be determined by comparing them to one another. The evaluation metrics that
will be examined are ROC AUC since they can be overly optimistic in situations when there are few samples
of the minority class and there is a badly imbalanced classification problem [16].
The evaluation metric focused on in this study is ROC AUC. This is a result of the dataset's
imbalanced number of classes, where there are a disproportionately large number of not fraudulent
transactions. Accuracy value for data with a uniform distribution is more significant than ROC AUC.
However, ROC AUC may be more significant in other circumstances, usually for extremely unbalanced data.
ROC-AUC is required to determine how well the model can distinguish between positive and negative
classes [17]. While accuracy is simply the proportion of accurate predictions, ROC AUC contrasts the
relationship between true positive rate and false positive rate [18].

Figure 2. The flow of creating the model

Int J Artif Intell, Vol. 13, No. 4, December 2024: 4273-4280

Int J Artif Intell ISSN: 2252-8938  4277

3. RESULTS AND DISCUSSION

The findings of the technique previously outlined in section 3 are summarized in Table 1. Based on
their ROC AUC value of 88.69%, random forest classifier using PSO as their optimization for
hyperparameter tuning produces the greatest results out of all machine learning methods as it is shown in
Figure 3. The top machine learning method, however, according to test accuracy, is XG-boost classifier,
which has a 97.28% accuracy rate.

Table 1. Summary ROC AUC value and accuracy with and without hyperparameter tuning
Machine learning models ROC AUC (%) Train accuracy (%) Test accuracy (%)
Random forest classifier Default 88.38 100.00 97.27
PSO 88.69 98.88 97.22
GA 87.73 97.06 96.45
XG-Boost classifier Default 87.53 97.96 97.21
PSO 87.27 98.43 97.21
GA 88.33 98.25 97.28
Light gradient boost machine Default 87.04 96.45 95.72
PSO 87.64 98.36 97.10
GA 84.99 97.01 96.76

Figure 3. Line graph of ROC AUC value comparison

Here are the findings after each machine learning model's hyperparameters were tuned using PSO
and GA techniques, with the ROC AUC calculated appropriately. Table 2 shows the set of best
hyperparameters that represent the ideal model configuration achieved according to each strategy and, as a
result, the highest accuracy, as well as the corresponding score (ROC AUC) and tuning techniques for
hyperparameters. It should be noted that while default settings produce excellent results for certain machine
learning models and datasets, this may not be circumstance when the model and/or data source are changed
[19]. According to our dataset, setting the default value for the 'learning rate' in the light gradient boost
machine using PSO to 0,1 yielded a ROC AUC value of 87,04%; however, setting this hyperparameter to
0.36 yielded a higher ROC AUC value of 87,64%.
According to the confusion matrix in Figure 4, the random forest classifier model with GA as its
hyperparameter optimization method correctly predicts the majority of fraud transactions (1,731 data). This
result is different from the best ROC AUC value achieved by the random forest classifier model with PSO as
its hyperparameter optimization method. This is due to the number of errors in predicting transactions that
are not fraud but predicted fraud is much more than the random forest classifier model with PSO as its
hyperparameter optimization with a difference of 903 data. The model with the best ROC AUC also managed
to predict fraudulent transactions correctly, namely as much as 1,575 data. The error in predicting fraudulent
transactions is also small, which is only 775 data. Therefore, the random forest classifier model with PSO as
its hyperparameter optimization method gets the best ROC AUC value in predicting fraudulent transactions
PSO outperforms GAs in some cases, and vice versa, implying that the two methods traverse the
problem in different ways. PSO successfully outperforms GA from ROC AUC result using random forest
classifier and light gradient boost machine learning model. On the other hand, GA outperforms PSO from
ROC AUC value using the XG-Boost classifier model. The findings of GA and PSO implementations in
fraud detection prove that the PSO algorithm outperforms the GA algorithm in both accuracy and iteration in
determining the best solution. PSO's key advantages over mathematical algorithms and other heuristic
optimization methods are its simple concept, straightforward implementation, adaptability to control
parameters, and computation time [20].

Elevating fraud detection: machine learning models with computational intelligence … (Cheryl Angelica)
4278  ISSN: 2252-8938

Table 2. Comparison of the best hyperparameter value in each machine learning model
HP approach Machine learning model ROC AUC (%) Best hyperparameters
PSO Random forest 88.69 'max_depth':37
'n_estimators':78
'max_features': 37
'min_samples_leaf': 6
'min_samples_split':5
XG-Boost 87.27 'max_depth': 7
'learning_rate': 0.36
'scale_pos_weight': 2
'gamma': 5
'reg_lambda': 1
'subsample': 0.76
'colsample_bytree': 0.29
Light gradient boost machine 87.64 'max_depth': -1,
'learning_rate': 0.36
'max_bin': 460
'min_data_in_leaf': 36
'feature_fraction': 0.98
GA Random forest 87.73 'max_depth':39
'n_estimators':27
'max_features': 8
'min_samples_leaf': 6
'min_samples_split': 7
XG-Boost 88.33 'max_depth': 10
'learning_rate': 0.11
'scale_pos_weight': 1
'gamma': 4
'reg_lambda': 0
'subsample': 0.91
'colsample_bytree': 0.17
Light gradient boost machine 84.99 ‘max_depth': -1
'learning_rate': 0.27
'max_bin': 52
'min_data_in_leaf': 10
'feature_fraction': 0.04

Figure 4. Confusion matrix comparison of machine learning models

Int J Artif Intell, Vol. 13, No. 4, December 2024: 4273-4280

Int J Artif Intell ISSN: 2252-8938  4279

Both GA and PSO employ population-based methodologies [21], sharing several characteristics.
While GA is recognized for discovering global optimal solutions [22], this research highlights a time
advantage for PSO over GA. Computational efficiency favors PSO, as evidenced by its lower computational
cost due to more apparent convergence [23] and a simpler parameter setup [24]. Unlike GA, PSO achieves a
steady state in locating.
Optimal solutions, eliminating the need for population destruction techniques [5]. Additionally,
PSO's simplicity in implementation and ease of parameter tuning makes it particularly appealing for
computationally constrained applications. This facilitates faster deployment and requires less fine-tuning
compared to GA [25].

4. CONCLUSION
Since today's scammers are professional rings with ample resources and modern equipment, unlike
in the early years of the internet, e-commerce fraud is growing swiftly. Many machines learning models have
been created for the purpose of recognizing fraud and have proven to be capable of saving billions of dollars
per year. It is faster, more scalable, efficient, and precise than humans at spotting insignificant trends or
illogical patterns. In this work, a comparison analysis has been conducted to identify the top machine
learning algorithms. XG-boost, random forest classifier, and light gradient boost machine models of machine
learning are employed. Metaheuristic optimization methods like GA and PSO have been applied as a
hyperparameter tuning strategy to improve the quality of machine learning models. Based on our experiment,
in contrast to GA, PSO provides a steady condition for finding the best solution. Without the use of
population elimination methods, PSO can get the best outcomes in various situations. Unlike GAs, which
must eliminate populations after they reach a specific saturation point in order to increase accuracy. PSO also
has faster computational time than GA. The findings indicate that random forest classifier provided the
highest ROC AUC value both before and after the hyperparameter tuning process, with a score of 88.69%
attained while utilizing PSO.

REFERENCES
[1] H. Zhou, G. Sun, S. Fu, L. Wang, J. Hu, and Y. Gao, “Internet financial fraud detection based on a distributed big data approach
with Node2vec,” IEEE Access, vol. 9, pp. 43378–43386, 2021, doi: 10.1109/ACCESS.2021.3062467.
[2] J. Nanduri, Y. Jia, A. Oka, J. Beaver, and Y. W. Liu, “Microsoft uses machine learning and optimization to reduce e-commerce
fraud,” Interfaces, vol. 50, no. 1, pp. 64–79, 2020, doi: 10.1287/inte.2019.1017.
[3] J. Shaji and D. Panchal, “Improved fraud detection in e-commerce transactions,” 2017 2nd International Conference on
Communication Systems, Computing and IT Applications, CSCITA 2017 - Proceedings, pp. 121–126, 2017, doi:
10.1109/CSCITA.2017.8066537.
[4] S. B. E. Raj and A. Annie Portia, “Analysis on credit card fraud detection methods,” 2011 International Conference on Computer,
Communication and Electrical Technology, ICCCET 2011, pp. 152–156, 2011, doi: 10.1109/ICCCET.2011.5762457.
[5] F. D. Wihartiko, H. Wijayanti, and F. Virgantari, “Performance comparison of genetic algorithms and particle swarm optimization
for model integer programming bus timetabling problem,” IOP Conference Series: Materials Science and Engineering, vol. 332,
no. 1, 2018, doi: 10.1088/1757-899X/332/1/012020.
[6] S. Khatri, A. Arora, and A. P. Agrawal, “Supervised machine learning algorithms for credit card fraud detection: A comparison,”
Proceedings of the Confluence 2020 - 10th International Conference on Cloud Computing, Data Science and Engineering, pp.
680–683, 2020, doi: 10.1109/Confluence47617.2020.9057851.
[7] R. Sailusha, V. Gnaneswar, R. Ramesh, and G. R. Rao, “Credit card fraud detection using machine learning,” Proceedings of the
International Conference on Intelligent Computing and Control Systems, ICICCS 2020, pp. 1264–1270, 2020, doi:
10.1109/ICICCS48265.2020.9121114.
[8] B. Youssef, F. Bouchra, and O. Brahim, “Rules extraction and deep learning for e-commerce fraud detection,” Colloquium in
Information Science and Technology, CIST, vol. 2020, pp. 145–150, 2020, doi: 10.1109/CiSt49399.2021.9357066.
[9] U. Porwal and S. Mukund, “Credit card fraud detection in E-commerce,” Proceedings - 2019 18th IEEE International Conference
on Trust, Security and Privacy in Computing and Communications/13th IEEE International Conference on Big Data Science and
Engineering, TrustCom/BigDataSE 2019, pp. 280–287, 2019, doi: 10.1109/TrustCom/BigDataSE.2019.00045.
[10] Q. Guo et al., “Securing the deep fraud detector in large-scale e-commerce platform via adversarial machine learning approach,”
The Web Conference 2019 - Proceedings of the World Wide Web Conference, WWW 2019, pp. 616–626, 2019, doi:
10.1145/3308558.3313533.
[11] K. Potdar, T. S. Pardawala, and C. D. Pai, “A comparative study of categorical variable encoding techniques for neural network
classifiers,” International Journal of Computer Applications, vol. 175, no. 4, pp. 7–9, 2017, doi: 10.5120/ijca2017915495.
[12] H. Liu, M. Zhou, and Q. Liu, “An embedded feature selection method for imbalanced data classification,” IEEE/CAA Journal of
Automatica Sinica, vol. 6, no. 3, pp. 703–715, 2019, doi: 10.1109/JAS.2019.1911447.
[13] A. Fernández, S. García, F. Herrera, and N. V. Chawla, “SMOTE for learning from imbalanced data: progress and challenges,
marking the 15-year anniversary,” Journal of Artificial Intelligence Research, vol. 61, pp. 863–905, 2018, doi:
10.1613/jair.1.11192.
[14] T. Friedrich, T. Kötzing, M. S. Krejca, and A. M. Sutton, “The compact genetic algorithm is efficient under extreme Gaussian
noise,” IEEE Transactions on Evolutionary Computation, vol. 21, no. 3, pp. 477–490, 2017, doi: 10.1109/TEVC.2016.2613739.
[15] H. T. Rauf, U. Shoaib, M. I. Lali, M. Alhaisoni, M. N. Irfan, and M. A. Khan, “Particle swarm optimization with probability
sequence for global optimization,” IEEE Access, vol. 8, pp. 110535–110549, 2020, doi: 10.1109/ACCESS.2020.3002725.
[16] A. J. Bowers and X. Zhou, “Receiver operating characteristic (ROC) area under the curve (AUC): a diagnostic measure for

Elevating fraud detection: machine learning models with computational intelligence … (Cheryl Angelica)
4280  ISSN: 2252-8938

evaluating the accuracy of predictors of education outcomes,” Journal of Education for Students Placed at Risk, vol. 24, no. 1, pp.
20–46, 2019, doi: 10.1080/10824669.2018.1523734.
[17] A. M. Carrington et al., “Deep ROC analysis and AUC as balanced average accuracy, for improved classifier selection, audit and
explanation,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 1, pp. 329–341, 2023, doi:
10.1109/TPAMI.2022.3145392.
[18] W. Deng, Z. Huang, J. Zhang, and J. Xu, “A data mining-based system for transaction fraud detection,” 2021 IEEE International
Conference on Consumer Electronics and Computer Engineering, ICCECE 2021, pp. 542–545, 2021, doi:
10.1109/ICCECE51280.2021.9342376.
[19] E. Elgeldawi, A. Sayed, A. R. Galal, and A. M. Zaki, “Hyperparameter tuning for machine learning algorithms used for arabic
sentiment analysis,” Informatics, vol. 8, no. 4, 2021, doi: 10.3390/informatics8040079.
[20] M. Issa, A. E. Hassanien, D. Oliva, A. Helmi, I. Ziedan, and A. Alzohairy, “ASCA-PSO: Adaptive sine cosine optimization
algorithm integrated with particle swarm for pairwise local sequence alignment,” Expert Systems with Applications, vol. 99, pp.
56–70, 2018, doi: 10.1016/j.eswa.2018.01.019.
[21] S. T. Li, B. Zhang, S. J. Xu, and Y. H. Zhong, “Back-analysis of pavement thickness based on PSO-GA hybrid algorithms,” IOP
Conference Series: Earth and Environmental Science, vol. 252, no. 5, 2019, doi: 10.1088/1755-1315/252/5/052066.
[22] A. Yazdanpanah, A. Rezaei, H. Mahdiyar, and A. Kalantariasl, “Development of an efficient hybrid ga-pso approach applicable
for well placement optimization,” Advances in Geo-Energy Research, vol. 3, no. 4, pp. 365–374, 2019, doi:
10.26804/ager.2019.04.03.
[23] Y. Ding, W. Zhang, L. Yu, and K. Lu, “The accuracy and efficiency of GA and PSO optimization schemes on estimating reaction
kinetic parameters of biomass pyrolysis,” Energy, vol. 176, pp. 582–588, 2019, doi: 10.1016/j.energy.2019.04.030.
[24] S. M. Almufti, A. Y. Zebari, and H. K. Omer, “A comparative study of particle swarm optimization and genetic algorithm,”
Journal of Advanced Computer Science & Technology, vol. 8, no. 2, pp. 40–45, 2019, doi: 10.14419/jacst.v8i2.29401.
[25] A. A. Karim, N. A. M. Isa, and W. H. Lim, “Modified particle swarm optimization with effective guides,” IEEE Access, vol. 8,
pp. 188699–188725, 2020, doi: 10.1109/ACCESS.2020.3030950.

BIOGRAPHIES OF AUTHORS

Cheryl Angelica is a master’s student at the School of Computer Science, Bina

Nusantara University, Indonesia. She entered college in 2019 and completed undergraduate
studies majoring in Information Technology (IT) in 2022. She then immediately takes a
master’s degree with the same major and is currently a last year student. Her research interests
are primarily in the area of artificial intelligence application, especially in machine learning,
deep learning, and data science. She can be contacted at email: [email protected].

Charleen is a dedicated college student in the Bina Nusantara University

Graduate Program (BGP). Pursuing a degree in Computer Science, she has actively
contributed to this paper through writing this paper. Showcasing a strong foundation in
computer vision. With a passion for research and a well-rounded approach to education, she
aspires to make significant future contributions to the field of computer science. She is
currently in her 9th semester of her college year at Bina Nusantara University, Jakarta 11480,
Indonesia. She can be contacted at email: [email protected].

Antoni Wibowo (M’12) received my first degree of Applied Mathematics in

1995 and master degree of Computer Science in 2000. In 2003, he awarded a Japanese
Government Scholarship (Monbukagakusho) to attend Master and Ph.D. programs at Systems
and Information Engineering in University of Tsukuba-Japan. He completed the second
master’s degree in 2006 and Ph.D. degree in 2009, respectively. His Ph.D. research focused on
machine learning, operations research, multivariate statistical analysis, and mathematical
programming, especially in developing nonlinear robust regressions using statistical learning
theory. He has worked from 1997 to 2010 as a researcher in the Agency for the Assessment
and Application of Technology–Indonesia. From April 2010–September 2014, he worked as a
senior lecturer in the Department of Computer Science–Faculty of Computing, and a
researcher in the Operation Business Intelligence (OBI) Research Group, Universiti Teknologi
Malaysia (UTM)–Malaysia. From October 2014–October 2016, he was an Associate Professor
at Department of Decision Sciences, School of Quantitative Sciences in Universiti Utara
Malaysia (UUM). He is currently working at Binus Graduate Program (Master in Computer
Science) in Bina Nusantara University-Indonesia as a specialist lecturer and continues his
research activities. He can be contacted at email: [email protected].

Int J Artif Intell, Vol. 13, No. 4, December 2024: 4273-4280