Global Transitions Proceedings 2 (2021) 35–41

Contents lists available at ScienceDirect

Global Transitions Proceedings

journal homepage: www.elsevier.com/locate/gltp

Credit card fraud detection using artificial neural network

Asha RB∗, Suresh Kumar KR
Department of ISE MSRIT, Bengaluru, Karnataka, India

a r t i c l e i n f o a b s t r a c t

Keywords: Frauds in credit card transactions are common today as most of us are using the credit card payment methods
Artificial neural network more frequently. This is due to the advancement of Technology and increase in online transaction resulting in
Credit card frauds causing huge financial loss. Therefore, there is need for effective methods to reduce the loss. In addition,
Fraud
fraudsters find ways to steal the credit card information of the user by sending fake SMS and calls, also through
k-Nearest Neighbor
masquerading attack, phishing attack and so on. This paper aims in using the multiple algorithms of Machine
Machine learning and support vector machine
learning such as support vector machine (SVM), k-nearest neighbor (Knn) and artificial neural network (ANN)
in predicting the occurrence of the fraud. Further, we conduct a differentiation of the accomplished supervised
machine learning and deep learning techniques to differentiate between fraud and non-fraud transactions.

1. Introduction pass the transaction details to the verification module where, it is clas-
sified under fraud and non-fraud categories. Any transaction that is
In recent years, as there is advancement of technology, most of them put under fraud category is rejected. Otherwise, the transaction gets
are using credit card for buying their needs so the frauds associated approved.
with it is also rising gradually. In the present world, almost all the en-
terprises from small to big industries are using the credit card as mode 1.2. Classifications of credit card frauds [13]
of payment. Credit card fraud is happening in all organization such as
appliances industry, automobile industry, banks and so on. Many of the 1 Application fraud: When a fraudster acquires the control over the
process like data mining, machine learning algorithmic approaches are application, steals the credentials of customer, and makes a fake ac-
applied to identify the fraud in the credit card transactions but did not count and then the transactions takes place.
get considerable result. Hence, there is a need of effective and efficient 2 Electronic or manual card imprints: In this form of fraud, the fraud-
algorithms to be developed that works significantly. we try to avoid the ster skims the information from the magnetic strip which is present
fraudster using our credit card before the transaction gets approved by on the card then uses the credentials and fraud transactions are car-
using artificial neural network algorithm and compared with few other ried out
machine learning algorithms. 3 Card not present: This is a type of credit card in which physical card
is not present during transaction
1.1. Overview of credit card fraud detection 4 Counterfeit card fraud: the fraud type in which the fraudster will
copies all the data from magnetic strip and real card looks like orig-
Fraud is an offensive activity, carried out by an unauthorized person inal card works as original card only. This card used for fraud.
by cheating innocent. Credit card fraud involves stealing the essential 5 Lost/stolen card: This type of fraud is due to loosing of the card by
credentials from the cardholder and using it unauthorized manner by the cardholder or by the stealing the card from the cardholder.
the fraudsters either by using phone calls or SMS. This fraud in credit 6 Card id theft: the type of fraud in which the id of the cardholder is
card may also happen using some software applications that are under stolen and fraud takes place.
the control of fraudsters. 7 Mail non-received card fraud: while issuing the credit card there will
The credit card fraud detection takes place as: the user or the cus- be procedure of sending a mail to the recipient, here fraud can occur
tomer enters the necessary credentials in order to make any transac- by defrauding the mail or phishing.
tion using credit card and the transaction should get approved only 8 Account Takeover: here the fraudster will take the complete control
upon being checked for ay fraud activity. For this to happen, we first of the account holder and make a fraud.

∗
Corresponding author.
E-mail addresses: [email protected] (A. RB), [email protected] (S.K. KR).

https://doi.org/10.1016/j.gltp.2021.01.006

Available online 23 January 2021

2666-285X/© 2021 The Authors. Publishing Services by Elsevier B.V. on behalf of KeAi Communications Co. Ltd. This is an open access article under the CC
BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
A. RB and S.K. KR Global Transitions Proceedings 2 (2021) 35–41

9 Fake fraud in website: fraudster will introduce a malicious code

which does their work in the website
10 Merchant collision: In this fraud type, cardholder details are shared
third party or the fraudster by merchants without cardholder autho-
rization.

The fraud in credit card transaction occurs when the stealer uses
the other person card without authorization of the respective person by
stealing the necessary information like PIN, password and other creden-
tials with or without the physical card. Using fraud detection module
involving machine learning and deep learning, we can find out whether Fig. 1. Architecture of Artificial neural network.
the upcoming transaction is fraud and legitimate.
Machine Learning is the treading and most used technology because
of its various applications and less time consumption, more accurate in ANN is biologically inspired by human brain. The neurons are inter-
result. Machine learning is a technology that deals with the algorithm, connected in the human brain like the same nodes are interconnected
which provides the computer, a capability to study and advance through in artificial neural network. Fig. 1 depicts the structure of ANN with
experience without being explicitly programmed. Machine learning has input, output and hidden layers. Inputs are x1, x2…Xn and output is
application in multiple fields. For example, medical, diagnosis, regres- y. w1…wn are the weights associated with inputs x1…xn respectively.
sion etc. Machine learning involves the combination of algorithm and There are 15 hidden layers used in this neural network. The activation
statically models which allow computer to perform the task without function used in our credit card fraud detection model is RELU.
hard coding then a model is build through a training data and then
it is tested on the trained model. 2. Related work
Deep learning is a part of machine learning techniques that makes
use of neural networks. Some of methods that come under deep learn- Some of the related study made by various researchers is described
ing are artificial neural network, Convolution neural network, autoen- in this section.
coders, recurrent neural networks, restricted Boltzmann machine etc. Altab Althar Taha and Sareef Jameel Malbery described that up-
Deep learning makes uses of neural networks, which resembles the hu- gradation in e-commerce and communication technology have made
man brain in processing the data and making the decisions. credit card usage more popular way of payment and the fraud associ-
ated with transactions is also increasing. They have used the optimized
1.3. Problem statement light gradient boosting machine, where Bayesian based hyper-parameter
optimization are combined to tune with parameter of light gradient
Now -a-days, most of them are using credit cards for buying the goods boosting machine (LightGBM). In this approach they used two set of
which are so much in need but can’t afford at the moment. In order to real world public dataset consisting of fraudulent and non-fraudulent
meet the needs credit cards are used and the fraud associated with it is transactions. Based on the comparison with other techniques, their pro-
also increasing so there is a need of developing a model that’s fit well posed system outperformed in terms of accuracy. The proposed system
and predicts at higher accuracy. produces the accuracy of 98.40%, area under receiver operating char-
acteristics curve (AUC) of 92.88%, Precision of 97.34% and F1-score of
56.95% [1].
1.4. Objectives
S. Makki, Z. Assaghir, Y. Taher, R. Haque, M. Hacid and H. Zeined-
dine research describes that the credit card fraud cause huge finan-
• The main objective of the research is to find a fraudulent transactions
cial loss. Most of the researchers have been working on this to pro-
in credit card transactions
vide an innovative ways to eradicate this loss and most of the available
• Comparison between the supervised learning and deep learning and
methods are costly, time consuming and labor incentive task. The au-
deep learning algorithm outperformed based on accuracy.
thors have found out that the imbalanced classification of dataset is the
main reason for the inaccurate results after many experimental studies.
1.5. Existing system These imbalance classifications consist of un-balanced dataset, which
caused the model to predict inaccurate and causes the financial loss.
The existing systems are carried out by considering machine learning Therefore, they have found that LR, C5.0 decision tree algorithm, SVM
algorithms like Support Vector Machine, Naïve Bayes, k-Nearest Neigh- and ANN are best algorithm based on accuracy, AUCPR and sensitiv-
bor and so on and some of them used random dataset. Very few have ity. They have used the balanced dataset in order to train these models
used artificial neural network for credit card fraud detection. [2].
C. Jiang, J. Song, G. Liu, L. Zheng and W. Luan proposed a novel
1.6. Proposed system process with multiple stages. First they collect the transactions made
by card holder, then based on the behavioural patterns transactions are
The Proposed system uses the Artificial Neural Network to find the aggregated, next the dataset is classified, further the model is trained
fraud in the credit card transactions. Performance is measured and ac- and finally the model is tested. If any abnormal behaviour arises then a
curacy is calculated based on prediction. And also classification algo- feedback is provided to system about the abnormal behaviour through
rithms such as Support vector machine and k-Nearest Neighbor are used feedback mechanism [3].
to build a credit card fraud detection model. We compare all the three Ishan Sohony, Rameshwar Pratap, and Ullas Nambiar proposed an
algorithms used in the experiment and made a decision that artificial ensemble learning approach for credit card fraud detection as the ratio
neural networks predicts well than system developed using support vec- of fraud to normal transaction is bit appropriate. They observed that
tor machine and k-nearest neighbor algorithms. The dataset used in the Random forest is best suited to provide a higher accuracy and neural
experiment consist of 31 attributes out of which 30 attributes consist of networks for detecting the fraud instances. They also experimented with
information related to name, age, account information and so on and the large real world credit card transactions. Ensemble learning is com-
last attribute give the outcome of the transaction in either 0 or 1. binations of Random forest and neural networks [4].

36
A. RB and S.K. KR Global Transitions Proceedings 2 (2021) 35–41

Phuong Hanh Tran, Kim Phuc Tran, Truong Thu Huong, Cédric
Heuchenne, Phuong HienTran, and Thi Minh Huong Le research de-
scribes that in past few years’ credit card fraud increased gradually.
Many methods performed using machine-learning algorithms to detect
the fraud transactions and block them. They introduced two new data
driven approaches, which uses the optimal anomaly technique for fraud
transaction in credit card transaction. The two ways are kernel param-
eter selection and T2 control chart [5].
Imane Sadgali, Nawal Sael, and Faouzia Benabbou work describes
that now-a-days banking transactions like online transactions, credit
card transactions and mobile transactions etc. are gaining popularity be-
cause all the people prefer digital and paperless transactions. Millions of
transactions carried out; all of them subjected to a type of fraud. Many
of the researchers have analyzed, designed and developed the model for
detecting the fraud using machine learning. They presented a compari-
son among the entire machine-learning algorithm to select which model
is best for fraud identification in card transaction [6]. Fig. 2. Complete architecture of proposed system.
Debachudamani Prusti and Santhnu Kumar Rath designed an appli-
cation with applied machine learning approaches such as Decision tree
(DT), k-nearest algorithm (kNN), Extreme learning machine (ELM), Mul- 3. Architectural design
tilayer perceptron (MLP) and support vector machine (SVM) to detect
the accuracy in fraud identification. They proposed a model by hybridiz- The system architecture will provide us a complete design of pro-
ing the DT, SVM and kNN techniques. They used two web-based proto- posed system. It explains about the implemented system architecture.
cols such as simple object access protocol (SOAP) and Representational The design of proposed system is shown in Fig. 2. The architecture
state transfer (REST) for efficient exchange of data across multiple het- depicts that first, user needs to register and then login into the system
erogeneous platforms. They compared five machine learning algorithm by entering the credentials like user name, password, and email-id and
results based on accuracy metric. SVM performed better than other al- phone number. Than it displays the algorithm selection windows where
gorithms by 81.63% but the hybrid system proposed by them had higher user need to select the algorithm. After selecting the algorithm, data
accuracy of 82.58% [7]. pre-processing takes place, which involves data cleansing, and normal-
Mohamad Zamini and Golamali montazar proposed unsupervised ization of dataset before fed into the actual model then splitting of data.
credit card Fraud detection system using autoencoders based cluster- Now the model is trained using Machine Learning algorithm and finally,
ing. They used three hidden layers and k means is used for clustering tested and predicted the end results whether transaction is fraud or not.
and tested on the European dataset which performed well compared to
other existing system [8].
4. Methods
Akila and Srinivasulu reddy proposed a misrepresentation location
framework with non-overlapped risk based bagging ensemble (NRBE)
This section explains about the implementation, which includes the
model to deal with the unevenness dataset and to keep away from the
algorithm used for implementation of proposed system.
noisiness contained in the transactions. They bagging model breaks all
In this paper, Implementations starts from loading the dataset. Than
the irregularity in dataset and non-vital nature. The sacking model is
data pre-processing carried out that includes data cleansing and normal-
reached out by pack of creation and danger based base student. The bag
izing the data. Dataset is splitted into two dataset as train data and test
creation eradicates the problem faced by imbalanced data and Naïve
data and model is trained and tested. Finally, system predicts whether
bayes destroys the issue brought about by noisiness created in the trans-
transaction is fraud or non-fraud.
actions. The proposed model has been beated with 5% in BCR and BER,
half of recall and 2x or 2.5x decreased expense to fraud detection by
utilizing the NBRE. It was distinguished that NRBE model best suits for 4.1. Programming language used
fraud detection and it is most appropriate for business dynamic tech-
nique [9]. In the implementation of proposed system, we used python as pro-
M. S. Kumar, V. Soundarya, S. Kavitha, E. S. Keerthika and E. Aswini gramming language. Python is beginner’s language, which provides var-
worked on project that made a model for detection of fraud in credit ious applications. In recent years, python had set the new trend because
card transactions using Random forest techniques. The random forest it is easy to use, interpreted, object-oriented, high-level, scripting lan-
algorithm (RFA) is a supervised machine learning techniques which uses guage. Python is one of the best languages for the implementing machine
the decision tree for classification of the credit card transactions and learning. It provides rich packages and libraries that used in machine
further performance is calculated using confusion matrix. The proposed learning.
system gives an accuracy of 90% [10].
Z. Li, G. Liu, S. Wang, S. Xuan and C. Jiang proposed fraud detec-
tion system via Kernel based supervised hashing (KSH). This KSH model 4.2. Packages and libraries used
based on approximate nearest neighbor idea. It is best suited for large
dataset with high dimension data. It is for the first time KSH is used for Some of the python library and packages used in proposed system
prediction, which performs better than other existing systems [11]. are as follows:
Pawan Kumar and fahad Iqbal made a survey on all techniques
used to detect the MasterCard fraud detection using machine-learning • Numpy
algorithms and evaluate the performance with the metrics. Tons of
analysis carried out over this domain. They say that there is a need Numpy is a python library. Abbreviation of Numpy is numerical
to use more efficient system that performs well at every situation python library. Numpy package is used for multidimensional arrays and
[12]. linear algebraic operations.

37
A. RB and S.K. KR Global Transitions Proceedings 2 (2021) 35–41

• Pandas Step 5: Calculating the fraud transactions and valid transactions,

then calculating the recall, precision and accuracy and stored in the
Pandas is a python library. Pandas is used for data analysis and data
respective locations
manipulation tool. It is used to read the dataset and load the dataset. It
Step 6: STOP
is fast, flexible when working with data.
• K-Nearest Neighbor
• Scikitlearn

A python package which is suitable for statistical model and machine Pseudocode:
learning models. A best suited python package for machine learning Step 1: START
modeling. Step 2: Loading of dataset pd.read.csv (csv file) # reads the csv file
and loads
• Keras Step 3: Cleaning and normalization of data
Keras is advanced stage of neural network application programming
• Normal = 0
interface (API). It is able of run on top of tensor flow. Keras is mainly
• Fraud = 1 # resampling
used while implementing deep learning algorithms such as CNN, RNN
• Data is scaled and normalized
because its user friendly, modularity, and easy to extensibility. It runs on
• Train_test_split() # splitting of dataset into train and test data
both CPU and GPU. In the experiment of finding the fraud or non fraud
credit card transaction we had used Keras along with backend running Step 4: Train the model then fit the trained model
tensor flow. This Keras along with tenor flow backend makes excellent
choice for training neural network architecture. • Trained the data using Knn classifier
• KNeighborsClassifier() # knn classifier which does classification of
• MySQL transactions
MySQL is database which is used for storage purpose. In the exper-
Step 5: Calculating the number of fraud, valid transactions and recall,
iment of fraud identification in card transaction we had used MySQL
precision and accuracy calculated.
for storing the user details namely user name, password, email-id and
Step 6: STOP
phone number. While entering into application, user needs to register
by providing the credential. These credentials are stored in database. • Artificial Neural Network (ANN):
Thereafter, user needs to login by giving username and password. The
application will validate the login and registered information than user Pseudocode:
is moved to next window. The ANN algorithm has two parts: Training part and testing part.
Training part:
• Tkinter Def ANN:
Tkinter is python library which is used for Graphical User interface Step 1: START
(GUI). It can be used on both Unix and Windows platform. We can create Step2: Loading and observing the dataset
it by importing Tkinter module then GUI is created and one or more
• pd.read.csv(.csv) # reads the dataset
widgets are added finally, called in loop.
• resampling of data
• StandardScaler() #scaling and normalization of data
4.3. Classification techniques
Step 3: Data pre-processing
List of algorithm used in implementation of our experiment are:
• Train_test_split() #Splitting of data
1 Support Vector Machine
2 K-nearest algorithm Step 4: Training the model
3 Artificial Neural Network
• Support Vector Machine: • Dense() #Adding data to activation function

Pseudocode: Step 5Analyzing the model

• Importing the necessary packages • Prediction of fraud is made and this trained data is stored .it can
Example: import pandas as pd used to test (training the model takes longer time so it is stored)

• def SVM Step 6: STOP

Testing part:
Step 1: START Def ANN
Step 2: Reading the dataset. pd.read.csv (file name) # reads the It is carried out similar way only difference is that the stored trained
dataset file model is used to test the data and classify it.
Step 3: Data cleaning and preprocessing of data

• Resampling the data as normal and fraud class i.e. normal = 0 and 5. Results and discussion
fraud =1 under
• Under sampling of data is done 5.1. Dataset
• Data is scaled (if any null value then eliminated) and normalized.
• Dataset is splitted into two set as train data and test data using split The proposed system makes use of the dataset downloaded from
() on training data is used to split the data. this website: www.kaggle.com . Dataset used is the transactions made
by customer in a European bank in the year 2013–14. It consist of 31
Step 4: Training the data using the SVM algorithm
columns, in which 30 columns are the features and the one class is the
• SVM classifier is called as classifier.predict () # which predicts target class which decides about whether the transaction is fraud or non-
whether transaction fraud or nonfraud. valid.

38
A. RB and S.K. KR Global Transitions Proceedings 2 (2021) 35–41

5.2. Evaluation measure Table 1

Represents the accuracy, recall and precision.
The end result is evaluated based on the confusion matrix and pre- Algorithms Accuracy Precision Recall
cision, recall and accuracy is calculated. It contains two classes: actual
SVM 0.9349 0.9743 0.8976
class and predicted class. The confusion metrics depends on these fea-
KNN 0.9982 0.7142 0.0393
tures: ANN 0.9992 0.8115 0.7619
True Positive: in which both the values positive that is 1.
True Negative: it is case where both values are negative that is 0.
False Positive: this is the case where true class is 0 and non-true
The Figs. 3–5 show the screenshot of results obtained by SVN, K-
class is 1.
NN AND ANN algorithm. It consists of count of fraud and non-fraud
False Negative: It is the case when actual class is 1 and non-true
instances, precision, recall and accuracy.
class is 0.
The Fig. 6 shows the accuracy performance measure of SVM, KNN
• Precision defined as follows: and ANN algorithms,This show that credit card fraud detection using ar-
tificial neural networks predicts at higher accuracy then Support vector
Precision = true positive / Actual result machine and k-nearest neighbor algorithms for fraud detection in credit
Precision = true positive/(true positive + false positive) card transactions.
• Recall defined as follows:
6. Conclusion
Recall = true positive / predicted result
Recall = true positive/(true positive + false negative)
In this research, we have proposed a method to detect the fraud in
• Accuracy defined as: credit card transactions that is based on deep learning. We first com-
pare it with machine learning algorithms such as k-Nearest Neighbor,
Accuracy = (true positive + true negative)/ total Support vector machine etc. Finally we have used the neural network,
even though tough to train the model which would fit fine to model for
5.3. Result detecting a fraud in credit card Transactions. In our model, by using
an artificial neural network (ANN) which gives accuracy approximately
The Table 1 shows the results of the used algorithms on the perfor- equal to 100% is best suited for credit card fraud detection .It gives ac-
mance metrics such as accuracy, precision and recall. curacy more than that of the unsupervised learning algorithms. In this

Fig. 3. shows the precision, accuracy and recall of the

SVM classifier.

Fig. 4. shows the result of Performance metrics using

K-NN classifier.

39
A. RB and S.K. KR Global Transitions Proceedings 2 (2021) 35–41

Fig. 5. shows the results using ANN.

Fig. 6. represent the plot of accuracy obtained using

SVM, KNN and ANN.

research work, data pre-processing, normalization and under-sampling International Conference on E-Business and Application. Association for Computing
carried out to overcome the problems faced by using an imbalanced Machinery, New York, NY, USA, 2018, pp. 6–9, doi:10.1145/3194188.3194196.
[6] I. Sadgali, N. Sael, F. Benabbou, Fraud detection in credit card transaction using
dataset. neural networks, in: Proceedings of the 4th International Conference on Smart City
Applications (SCA ’19). Association for Computing Machinery, New York, NY, USA,
References 2019, pp. 1–4, doi:10.1145/3368756.3369082. Article 95.
[7] D. Prusti, S.K. Rath, Web service based credit card fraud detection by applying ma-
[1] A.A. Taha, S.J. Malebary, An intelligent approach to credit card fraud detection using chine learning techniques, in: Proceedings of the TENCON 2019 - 2019 IEEE Re-
an optimized light gradient boosting machine, IEEE Access 8 (2020) 25579–25587, gion 10 Conference (TENCON), Kochi, India, 2019, pp. 492–497, doi:10.1109/TEN-
doi:10.1109/ACCESS.2020.2971354. CON.2019.8929372.
[2] S. Makki, Z. Assaghir, Y. Taher, R. Haque, M. Hacid, H. Zeineddine, An experimental [8] M. Zamini, G. Montazer, Credit card fraud detection using autoencoders based clus-
study with imbalanced classification approaches for credit card fraud detection, IEEE tering, in: Proceedings of the 9th International Symposium on Telecommunications
Access 7 (2019) 93010–93022, doi:10.1109/ACCESS.2019.2927266. (IST), Tehran, Iran, 2018, pp. 486–491, doi:10.1109/ISTEL.2018.8661129.
[3] C. Jiang, J. Song, G. Liu, L. Zheng, W. Luan, Credit card fraud detection: a novel [9] S. Akila, U.S. Reddy, Credit card fraud detection using non-overlapped risk based
approach using aggregation strategy and feedback mechanism, IEEE Internet Things bagging ensemble (NRBE), in: Proceedings of the IEEE International Conference
J. 5 (5) (Oct. 2018) 3637–3647, doi:10.1109/JIOT.2018.2816007. on Computational Intelligence and Computing Research (ICCIC), Coimbatore, 2017,
[4] I. Sohony, R. Pratap, U. Nambiar, Ensemble learning for credit card fraud detection, pp. 1–4, doi:10.1109/ICCIC.2017.8524418.
in: Proceedings of the ACM India Joint International Conference on Data Science and [10] M.S. Kumar, V. Soundarya, S. Kavitha, E.S. Keerthika, E. Aswini, Credit card fraud
Management of Data Association for Computing Machinery, New York, NY, USA, detection using random forest algorithm, in: Proceedings of the 3rd International
2018, pp. 289–294, doi:10.1145/3152494.3156815. Conference on Computing and Communications Technologies (ICCCT), Chennai, In-
[5] P.H. Tran, K.P. Tran, T.T. Huong, C. Heuchenne, P. HienTran, T.M.H. Le, Real time dia, 2019, pp. 149–153, doi:10.1109/ICCCT2.2019.8824930.
data-driven approaches for credit card fraud detection, in: Proceedings of the 2018

40
A. RB and S.K. KR Global Transitions Proceedings 2 (2021) 35–41

[11] Z. Li, G. Liu, S. Wang, S. Xuan, C. Jiang, Credit card fraud detection via kernel- [12] P. Kumar, F. Iqbal, Credit card fraud identification using machine learning ap-
based supervised hashing, in: Proceedings of the IEEE SmartWorld, Ubiquitous In- proaches, in: Proceedings of the 1st International Conference on Innovations in Infor-
telligence & Computing, Advanced & Trusted Computing, Scalable Computing & mation and Communication Technology (ICIICT), CHENNAI, India, 2019, pp. 1–4,
Communications, Cloud & Big Data Computing, Internet of People and Smart City In- doi:10.1109/ICIICT1.2019.8741490.
novation (SmartWorld/SCALCOM/UIC/ATC/CBDCom/IOP/SCI), Guangzhou, 2018, [13] Y. Jain, N. Tiwari, S. Dubey, S. Jain, A comparative analysis of various credit card
pp. 1249–1254, doi:10.1109/SmartWorld.2018.00217. fraud detection techniques, Int. J. Recent Technol. Eng. 7 (2019) 402–407.

41