Content uploaded by Khadija Abdulsattar
Author content
All content in this area was uploaded by Khadija Abdulsattar on Jan 20, 2021
Content may be subject to copyright.
Fraudulent Transaction Detection in FinTech using
Machine Learning Algorithms
Khadija AbdulSattar
College of Information Technology
University of Bahrain
Sakhir, Bahrain
20124209@stu.uob.edu.bh
Mustafa Hammad
College of Information Technology
University of Bahrain
Sakhir, Bahrain
mhammad@uob.edu.bh
Abstract— With the advancement of e-commerce, online
transactions purchases using credit and debit cards have
drastically increased. This has caused a burst in credit and debit
card fraud and has become a profoundly significant global issue.
Fraud touches every area of our lives and is a growing concern
that effects both businesses and customers. As machine learning
techniques provide unique and efficient solutions, they are
applicable in various types of problems. Recently, machine
learning algorithms have been widely applied as a data mining
technique for classification problems. In this paper, a binary
classification problem is considered where a transaction can be
classified as either fraudulent or legitimate transaction. The goal
is to classify the transactions using five different machine
learning algorithms. The transaction dataset (Task1 and Task2)
is preprocessed, and then SGD, DT, RF, J48 and IBk machine
learning classifiers are applied. After applying the classifiers,
the results are compared to analyze which classifier performs
the best. Based on the experimental results, it is found that the
accuracy percentage of all the five classifiers for Task1 and
Task2 datasets is ranging between 97.78% to 98.1%, with no
major difference. As the dataset is highly imbalanced, the
kappa statistic value is also considered. For both datasets, the
RF classifier had the greatest value of kappa statistics, whereas
SGD and J48 had the lowest value for Task1 and Task2,
respectively. Other evaluation metrics were also considered for
evaluating the performance of the applied classifiers. Overall,
these classifiers achieved similar results for Task2 dataset. As
negative Kappa statics and MCC values were obtained, the SGD
classifier for Task2 dataset had the worst results in comparison.
Based on evaluation criteria such as Kappa statistic and MCC
values RF outperformed the others for both the datasets.
Keywords— classification; credit and debit cards transactions;
ML algorithms; imbalanced data; fraud detection
I. INTRODUCTION
In this digital-era of information, the number of
transactions being performed through online purchases using
credit and debit cards has drastically increased. Fraudulent
transaction detection of debit and credit cards is typically
considered as a binary classification problem, where each
transaction may either be classified as either legitimate or
fraudulent. Thus, fraud detection has become a deeply
relevant global issue. Therefore, many techniques have been
used by the FinTech industry for fraud detection [1-3].
As machine learning techniques provide unique and
efficient solutions, they are applicable in various types of
problems. There are a large number of research studies
conducted on machine learning as a popular method for
classification problems. Machine learning helps in revealing
scams in different domains, such as FinTech, Healthcare, and
eCommerce. There is great emphasis given in the research
studies on data mining and neural network for detecting
fraudulent transactions [4][5]. In [6], Bhatia et al. discussed a
few detection techniques used in the literature. These
techniques include a fusion approach, Bayesian and neural
networks, hidden Markov model, SVM, fuzzy Darwinian
detection, kNN, and Naïve Bayes.
Although, fraud detection has a long history, but due to
data deficiency and other challenges this area has not been
developed much. Due to privacy reasons, the real data of
banks customers’ transactions needs to be kept confidential,
and cannot be revealed. Thus, real transaction data is not
easily available for exploration. Moreover, the field names of
the dataset are changed for privacy reasons. Many challenges
are encountered during fraud detection. Puh and Brkić have
discussed some of these challenges in [7]. These challenges
include data deficiency, imbalanced data, behavioral
variation, cost sensitive problem and evaluation metric.
There are different kinds of frauds found in credit card
transactions. In [8], G. and G. Gupta explained a few amongst
them. These include identity theft, unauthorized purchases
using lost/stolen cards, ATO (Account TakeOver) i.e.
unauthorized usage of online accounts, and faking and
skimming/counterfeiting cards. They performed the fraud
detection in two phases, feature extraction and then
classification. According to [9], in 2018 identity theft fraud
reports, credit card fraud ranked the first. It is being the most
common and popular kind of identity theft. It was also stated
that credit card fraud increased by 18.4 percent in 2018 and is
still climbing [9]. Therefore, fraud detection and prevention
are profoundly a significant global issue.
In this paper, a binary classification problem is considered
where the target attribute will identify a transaction as either
fraudulent or legitimate. The goal is to classify the
transactions using different machine learning classifiers. The
transaction dataset will be preprocessed, and then five
different machine learning classifiers will be applied. After
applying the classifiers, the results will be compared to
analyze which classifier performs the best.
The sections covered of this paper are as follows. In
section II, the related work is given followed by section III
where the research methodology is briefly explained. Section
IV describes the machine learning algorithms applied in the
research. Then, the details of the dataset used, and its pre-
processing techniques are explained in section V. Section VI
gives all the details about the evaluation measures selected for
the performance evaluation of the applied machine learning
algorithms. The experimental results are given in section VII.
In section VIII, the paper ends with the conclusion and future
work.
II. RELATED WORK
Various techniques have been offered to detect fraudulent
transactions using neural networks. In [10], a fraud detection
method for credit cards was proposed by Ghosh and Reilly
which was based on neural networks. Their proposed system
was installed in a bank for fraud detection. Similar to [10],
Brause et al. [11] presented a credit card fraud prediction
2020 International Conference on Innovation and Intelligence for Informatics, Computing and Technologies
(3ICT)
978-1-7281-9673-2/20/$31.00 ©2020 IEEE
technique depending on neural networks. Aleskerov et al. [12]
proposed a technique called CARDWATCH. This is a system
used for similar purpose. It provides a graphical user interface
for different commercial databases. This system was based on
neural network model and provided very convincing fraud
detection rates. Another study [13], proposed a model for
fraud detection based on neural networks. This was a
confidence-based neural network. In this model, a technology
for fraud detection that ensured the accuracy and effectiveness
was introduced. A transaction was considered as fraudulent, if
it had sufficiently low confidence. A parallel Granular Neural
Network (GNN) was developed by Syeda et al. [14] for
detecting credit card fraud.
There are many studies about fraud detection prediction
using hybrid methods. In [15], Patidar and Sharma used
Artificial Neural Networks (ANN) along with the genetic
algorithm. Here, particularly back propagation neural network
was used in the algorithm. Another study [16], also used a
combination of techniques. Shen et al. [16] investigated the
efficiency of three techniques and then proposed a framework
to choose the best among those for detection. The three
techniques, decision tree, neural networks and Logistic
Regression (LR), were investigated for detecting fraud. In
[17], credit card fraud detection classification models based
on ANN and LR were implemented and used. It compared the
performance of ANN and LR with real data set. A novel
approach for credit card fraud detection was proposed by
Panigrahi et al. [18]. This proposed approach is a fusion
approach which used Dempster–Shafer theory and Bayesian
learning. They compared their approach with other methods.
It was found that the performance of their approach evidenced
very high positive impact in fraud detection for credit cards.
In [19], Şahin and Duman developed a technique based on
decision trees and SVM. They applied this technique for
detecting credit card fraud to a real data set and compared the
performance of the two techniques. In [3], hybrid methods
which used AdaBoost and majority voting methods were
tested on a credit card data set which was publicly available.
Then, the model was evaluated.
Moreover, many studies discussed Hidden Markov Model
(HMM) for detecting fraud. One such approach was
developed by Srivastava et al. [20]. Their model was trained
such that, if the transaction had sufficiently high probability,
it was considered as fraudulent. Simultaneously, it was
ensured that legitimate transactions were not rejected. They
compared their models with other techniques in the literature.
Likewise, in [21], Dhok and Bamnote presented HMM for
detecting credit card fraud during transactions. The proposed
HMM helped in achieving a high accuracy on fraud detection
while having low false alarm rate. Khan et al. [22] also used
HMM for detecting credit card fraud. The experimental results
were presented to prove their model’s effectiveness. In [23],
an Optimized Multiple Semi-HMM (OMSHMM) was
developed to optimize the model parameters. It effectively
detected fraud transactions. Furthermore, [24] and [25] also
used HMM for detecting credit card fraud.
Some related works also discussed approaches which were
not based on machine learning algorithms for fraud detection.
One such technique was based on association rules, applied by
Sánchez et al. [26]. They proposed its use for preventing fraud
and detecting unlawful credit card transactions. Their
proposed methodology was applied in some retail companies.
In [27], Quah et al. presented a model for detecting credit card
fraud in real-time. Their proposed approach used
computational intelligence. It made use of self-organizing
map to decipher possible fraud cases. It could also filter and
analyze customer behavior and spending patterns. Ganji and
Mannem [28] proposed a technique called Stream Outliner
Detection algorithm based on Reverse k-Nearest Neighbors
(SODRNN). Moreover, Chiu and Tsai [29] offered a mothed
for detecting fraud in credit card transactions based on web
services collaborative scheme.
From the previous studies, it is found that a variation of
machine learning classifiers has been widely applied for
detecting fraud in transactions. Most widely used machine
learning techniques were based on ANN, LR, SVM, Hidden
Markov Model, kNN and decision trees. The use of machine
learning techniques based on hybrid models were also applied.
In this research, machine learning algorithms RF, SGD,
decision table are also applied along with decision trees (J48),
and kNN (IBk) which were already applied in other works.
Findings of different ML classifiers are compared to evaluate
their performance. A number of evaluation measures are
considered in order to get a better insight of differ results in
comparison.
III. METHODOLOGY
In this experimental research, the transaction dataset called
UCSD Data Mining Contest 2009 Dataset is used. This
research is emphasized on the classification of transactions
into either fraudulent or legitimate. The dataset is
preprocessed before applying machine learning classifiers for
training and testing. Then the performance of different
machine learning algorithms is to be compared on this
classification problem using Weka tool. The machine learning
algorithms analyzed for the transaction dataset are briefly
explained in section IV and then the results of different
classifiers are to be analyzed and compared. The performance
of each classifier is evaluated by a number of classification
evaluation measures. Fig. 1 illustrates the model used for
fraudulent transaction detection and research design.
Fig. 1. Model for fraudulent transaction detection and research design
IV. MACHINE LEARINING ALGORITHMS USED
In this study, Weka tool is used as a data mining software
for evaluating the datasets. It offers several machine learning
algorithms for performing data exploration and data mining
tasks. Five machine learning algorithms are selected and the
datasets will be tested using these algorithms. These machine
learning algorithms are Stochastic Gradient Descent (SGD),
2020 International Conference on Innovation and Intelligence for Informatics, Computing and Technologies
(3ICT)
Random Forest (RF), Decision Table (DT), Decision Tree
(J48) and Nearest Neighbor algorithm - Instance Based-k
(IBk). Each classifier is briefly explained below.
• Stochastic Gradient Descent (SGD): It is an optimization
algorithm used for large-scale ML problems, such as
natural language processing, as well as, for discriminative
learning of linear classifiers [30].
• Random Forest (RF): It is a ML algorithm applied to
classification as well as regression problems. Decision
trees on data samples are generated. Then, by obtaining the
prediction from each generated tree, the best one is chosen
based on voting [31].
• Decision Table (DT): It is also a machine learning
classification model used for prediction. They are similar
to decision trees.
• Decision Tree (J48): It is a supervised learning technique.
J48 decision tree classifier is also known as C4.5. It is an
algorithm used to generate decision trees. It is used for
classification and also known as a statistical classifier [32].
• Nearest Neighbor algorithm - Instance Based-k (IBk):
kNN machine learning algorithms is based on supervised
learning technique. While classifying an instance using
IBk's KNN, its parameter is used for specifying the
number of nearest neighbors to use. Then, the outcome is
based on majority vote [33].
V. DATA SET AND PRE-PROCESSING
Various kinds of datasets for credit card transaction are
available with various transaction attributes. To analyze the
classification output of different machine learning algorithms,
UCSD Data Mining Contest 2009 Dataset [34] is selected.
This dataset is a real dataset of credit card transactions. Fraud
detection in credit card transactions is a binary classification
problem. Therefore, each credit card transaction belongs to
one of the two classes, legitimate or fraudulent. The dataset
contains two versions: easy (Task1) and hard (Task2)
versions. Both of these versions were used for the evaluation
of the classifiers’ output.
Table I illustrates fraudulent and legitimate transaction
distribution in Task1 and Task2 datasets. It is seen that there
is a huge number of fraudulent transactions as compared to
legitimate ones. This shows that these datasets are highly
imbalanced.
TABLE I. TASK1 AND TASK2 DATASET INSTANCES AND
ATTRIBUTES
Dataset details
Dataset version
Task1
Task2
Number of instances
94682
100000
Number of attributes (original)
20
20
Number of attributes (after pre-processing)
16
16
Number of fraudulent transactions
2094
2654
Number of legitimate transactions
92588
97346
These datasets were pre-processed by organizing the
original dataset, simplifying it and removing all uncommon
attributes between the two datasets. As mentioned in Table I,
both the original datasets had 20 attributes but after pre-
processing 16 attributes were used for analysis. The list of
attributes after pre-processing are amount, hour1, state1,
field1, field2, hour2, total, flag1, field4, indicator1, indicator2,
flag2, flag3, flag4, flag5, and Class. The fields amount, hour1,
hour2, state1, total and class have descriptive column name
representing corresponding values for each customer. All the
other fields field1, field2, flag1, field4, indicator1, indicator2,
flag2, flag3, flag4 and flag5 are anonymized, therefore, are
kept as they are. Due to privacy reasons, some attributes are
found to be representing the same information as labelled,
while all other fields are anonymized. Each version of Task1
and Task2 dataset is further divided into two datasets: the
training dataset and the testing dataset. The training dataset
has labelled class values, whereas the testing dataset is
unlabeled. Henceforth, the training dataset of both the
versions Task1 and Task2 are used.
VI. EVALUATION MEASURES
For evaluating the output of the machine learning
algorithms, well known evaluation measures are considered.
The performance of each classifier is evaluated by
classification evaluation measures. The evaluation measures
are selected based on their relevance to the problem.
Commonly applied evaluation measures in credit card fraud
detection problem for assessing the classifier’s performance
include correctly classified rate, fraud detection rate, Kappa
statistic, precision, recall, F-measure, RCC (Receiver
Operating Characteristic) area [35] and MCC (Matthews
correlation coefficient). The performance of each classifier is
evaluated using the below measures:
A. Confusion matrix
Confusion matrix as illustrated in Table II, is generated as
a part of the classifier output. It provides TP, TN, FP, and FN
values, which is True Positive, True Negative, False Positive
and False Negative, respectively. Here, positive class
indicates fraudulent and negative class indicates legitimate,
hence, the terms TP, TN, FP, and FN are defined as follows:
• TP: the number of fraudulent transactions classified as
fraudulent.
• TN: the number of legitimate transactions classified as
legitimate.
• FP: the number of legitimate transactions classified as
fraudulent.
• FN: the number of fraudulent transactions classified as
legitimate.
TABLE II. 2 X 2 CONFUSION MATRIX
Predicted
Actual
Fraudulent
Legitimate
Fraudulent
TP
FP
Legitimate
FN
TN
B. Accuracy
Accuracy [36] is the ratio of the number of correctly
classified instances to the total number of classified instances.
If the value of the ratio is 1, it indicates best accuracy, whereas
the value 0 indicates the worst. It is computed as in (1).
()
C. Precision
Precision [36] is the ratio of the number of correctly
classified positive instances to the total number of positive
classified instances. If the value of the ratio is 1, it indicates
best precision, whereas, the value 0 indicates the worst. It is
computed as in (2).
2020 International Conference on Innovation and Intelligence for Informatics, Computing and Technologies
(3ICT)
D. Recall
Recall [28] is the ratio of the number of classified positive
instances to the total number of positive instances. If the value
of the ratio is 1, it indicates best recall, whereas, the value 0
indicates the worst. It is computed as in (3).
()
E. F-measure
F-measure [36] is the weighted harmonic mean of
precision and recall of the test, calculated based on the values
of recall and precision measures combined in one. It is
computed as in (4).
()
F. Kappa statistics
Kappa statistics is an evaluation measure which compares
an observed accuracy with an expected accuracy. If the value
of Kappa statistic is greater than 0 it indicates better
classification. The closer the value to 1 it interprets a better
agreement beyond chance. It is computed as in (5).
()
G. MAE and RMSE
MAE and RMSE are also used for evaluating the
performance of a machine learning prediction model. They
are used to measure the difference between the values
predicted by the model and the actual observed values. If the
actual value is a and the predicted value is p then MAE and
RMSE are calculated as in (6) and (7).
()
()
H. MCC
Matthews Correlation Coefficient (MCC) [37] is used for
the classifier’s performance evaluation measures for binary
classification problem. It uses TP, FP, TN and FN values. This
is one the measures considered for an imbalanced dataset,
such as credit card fraud detection dataset. It is a correlation
coefficient between the observed and predicted two-class
classifications. Its value ranges between -1 to +1, where +1
indicates best prediction, and -1 indicates otherwise. The
value 0 indicates no better prediction than chance. It is defined
as in (8).
()
VII. EXPERIMENTAL RESULTS
When testing the dataset using the above-mentioned
classifiers, cross-validation (folds=10) as test option is used.
In this test mode, the data in datasets are automatically split
between training data and testing data. In each cross-
validation fold, 90% of the data is taken for training the model
while the remaining 10% of the data is used for testing. Thus,
this process is repeated in each fold.
Fig.2 and Fig. 3 show the percentage of correctly classified
instances for both datasets Task1 and Task2 ranges between
97.56% to 98.1%, indicating most of the instances are
classified correctly. The number of correctly classified
instances indicates the sum of TP and TN values. The number
of incorrectly classified instances indicates the sum of FP and
FN values. These TP, TN, FP and FN values represent
confusion matrices which are given in Table III.
Fig. 2. Comparison between correctly and incorrectly classified
transactions of each classifier on Task1 dataset
Fig. 3. Comparison between correctly and incorrectly classified
transactions of each classifier on Task2 dataset
As mentioned earlier, these datasets are highly
imbalanced. Therefore, some other evaluation measures are
also needed to be taken under consideration. The accuracy
percentage alone cannot be used to conclude the best
classifier. One important measure in such datasets is Kappa
statistic, if the value is greater than 0 it indicates better
classification. The closer the value to 1 interprets a better
agreement beyond chance. For Task1 the Kappa statistic
values ranges from -0.0001 (SGD classifier) to 0.3381 (RF),
indicating SGD performed worst and RF performed the best.
Unlike the Task1 Kappa statistic results, Task2 all positive
values indicating better performance, the values ranging from
0.3889 (DT) to 0.4657 (RF). Hence, RF outperforms other
classifiers in both Task1 and Task2 datasets. Fig. 4 and Fig. 5
show the comparison between Kappa statistic, MAE and
RMSE of each classifier SGD, DT, RF, J48 and IBk on Task1
and Task2 datasets, respectively.
The confusion matrix generated as a part of each classifier
output is shown in Table III. It further explains the generated
results, providing TP, TN, FP and FN values for Task1 and
Task2 datasets.
()
2020 International Conference on Innovation and Intelligence for Informatics, Computing and Technologies
(3ICT)
Fig. 4. Comparison between Kappa statistic, MAE and RMSE of each
classifier on Task1 dataset
Fig. 5. Comparison between Kappa statistic, MAE and RMSE of each
classifier on Task2 dataset
TABLE III. CONFUSION MATRIX OF 5 CLASSIFIERS’ OUTPUT
FOR TASK1 AND TASK2
Classifier
Predicted
Actual
Task1
Task2
Fraudulent
Legitimate
Fraudulent
Legitimate
SGD
Fraudulent
92583
5
97185
161
Legitimate
2094
0
1947
707
DT
Fraudulent
92485
103
97177
169
Legitimate
1827
267
1955
699
RF
Fraudulent
92413
175
96970
376
Legitimate
1621
473
1710
944
J48
Fraudulent
92497
91
97187
159
Legitimate
1800
294
1950
704
IBk
Fraudulent
92124
464
96617
729
Legitimate
1587
507
1710
944
Table IV show the detailed accuracy by class of all the 5
classifiers used for Task1 and Task2 datasets. Precision,
Recall, F-Measure and MCC are well-known evaluation
measures. For each classifier, the values the row indicates the
weighted average value of the evaluation measures. The
weighted average values of precision, recall and F-measure of
all the classifiers for both Task1 and Task2 gives strong results
with the values ranging from 0.956 to 0.981. The MCC values
of Task1 varies for each classifier, with the lowest being -
0.001 (SGD classifier) and highest being 0.4 (RF). Based on
the MCC values SGD performed worst and RF performed
best. Unlike Task1 values of MCC, the Task2 values of MCC
are similar for all classifiers ranging between 0.436 to 0.495.
Hence, all the classifiers performed almost the same and are
positive values. To be precise the highest value was achieved
using RF classifier i.e. 0.495.
TABLE IV. WEIGHTED AVERAGE VALUES OF PRECISION,
RECALL F-MEASURE AND MCC OF 5 CLASSIFIERS’ OUTPUT FOR
TASK1 AND TASK2
Weighted Average
for Classifier
Task1
Task2
Precision
Recall
F-Measure
MCC
Precision
Recall
F-Measure
MCC
SGD
0.956
0.978
0.967
-0.001
0.976
0.979
0.974
0.459
DT
0.975
0.980
0.973
0.298
0.976
0.979
0.973
0.453
RF
0.977
0.981
0.976
0.400
0.976
0.979
0.976
0.495
J48
0.976
0.980
0.973
0.322
0.976
0.979
0.974
0.458
IBk
0.973
0.978
0.974
0.346
0.972
0.976
0.973
0.436
VIII. CONCLUSION AND FUTURE WORK
With the popularity and upsurge of online transactions,
credit card fraud detection has become an essential need and
requirement. In this paper a binary classification problem was
considered where the target attribute identified a transaction
as either fraudulent or legitimate. The goal was to classify the
transactions using different machine learning algorithms. The
machine learning algorithms applied were SGD, RF, DT, J48
and IBk. For conducting this research, the data mining tool
called Weka was used and the datasets were evaluated. For
analyzing the classification output of different machine
learning algorithms, UCSD Data Mining Contest 2009
Dataset was selected. The transaction dataset was pre-
processed, and then the selected machine learning classifiers
were applied. After applying the classifiers, the results were
compared to analyze which classifier performs the best.
Accuracy, precision, recall, F-measure, Kappa statistics,
MAE, RMSE, MCC and confusion matrix were used. As the
datasets that were used are highly imbalanced, therefore, some
other evaluation measures were also needed to be taken under
consideration. The accuracy percentage alone cannot be used
to conclude the best classifier. Based on the overall
experimental results, it is found that for Task1 dataset SGD
performed worst and RF performed the best. For Task2
dataset, all the applied classifiers had similar results as per the
accuracy. The weighted average values of precision. recall and
F-measure of all the classifiers for both Task1 and Task2 gave
strong results with the values ranging from 0.956 to 0.981.
Based on evaluation criteria such as Kappa statistic and MCC
values RF outperformed the others.
This research was limited to classify fraudulent transaction
from real dataset that has already happened. Along with fraud
detection, fraud prevention is of vital importance. To avoid the
damage to happen, it is crucial to prevent such fraudulent
transactions in a timely fashion before causing any damage. In
the future, this study may further be extended to explore more
credit card fraud detections using balanced dataset and by
applying feature selection techniques.
REFERENCES
[1] T. Hastie, R. Tibshirani and J. Friedman, The elements of statistical
learning. New York, NY: Springer, 2009.
[2] D. Jensen, "Prospective assessment of ai technologies for fraud
detection: A case study", in AAAI Workshop on AI Approaches to
Fraud Detection and Risk Management, 1997, pp. 34-38.
[3] K. Randhawa, C. Loo, M. Seera, C. Lim and A. Nandi, "Credit Card
Fraud Detection Using AdaBoost and Majority Voting", IEEE Access,
2020 International Conference on Innovation and Intelligence for Informatics, Computing and Technologies
(3ICT)
vol. 6, pp. 14277-14284, 2018. Available:
10.1109/access.2018.2806420 [Accessed 22 February 2020].
[4] S. Benson Edwin Raj and A. Annie Portia, "Analysis on credit card
fraud detection methods", 2011 International Conference on
Computer, Communication and Electrical Technology (ICCCET),
2011. Available: 10.1109/icccet.2011.5762457 [Accessed 22 February
2020].
[5] S. Maes, K. Tuyls, B. Vanschoenwinkel and B. Manderick, "Credit
Card Fraud Detection Using Bayesian and Neural Networks", First
International NAISO Congress on Neuro Fuzzy Technologies, Havana,
Cuba, 2002. [Accessed 22 February 2020].
[6] S. Bhatia, R. Bajaj and S. Hazari, "Analysis of Credit Card Fraud
Detection Techniques", International Journal of Science and Research
(IJSR), vol. 5, no. 3, pp. 1302-1307, 2020. [Accessed 22 February
2020].
[7] M. Puh and L. Brkic, "Detecting Credit Card Fraud Using Selected
Machine Learning Algorithms", 2019 42nd International Convention
on Information and Communication Technology, Electronics and
Microelectronics (MIPRO), 2019. Available:
10.23919/mipro.2019.8757212 [Accessed 22 February 2020].
[8] G. Gupta, "Analysis of Various Credit Card Fraud Detection
Techniques", International Journal of Computer Science and Mobile
Computing, vol. 8, no. 5, pp. 8-14, 2019. [Accessed 22 February 2020].
[9] "Credit Card Fraud Statistics", Shift Credit Card Processing, 2020.
[Online]. Available: https://shiftprocessing.com/credit-card-fraud-
statistics/. [Accessed: 19- Oct- 2020].
[10] Ghosh and Reilly, "Credit card fraud detection with a neural-network",
Proceedings of the Twenty-Seventh Hawaii International Conference
on System Sciences HICSS-94, 1994. Available:
10.1109/hicss.1994.323314 [Accessed 29 February 2020].
[11] R. Brause, T. Langsdorf and M. Hepp, "Neural data mining for credit
card fraud detection", Proceedings 11th International Conference on
Tools with Artificial Intelligence. Available: 10.1109/tai.1999.809773
[Accessed 29 February 2020].
[12] E. Aleskerov, B. Freisleben and B. Rao, "CARDWATCH: a neural
network based database mining system for credit card fraud detection",
Proceedings of the IEEE/IAFE 1997 Computational Intelligence for
Financial Engineering (CIFEr). Available: 10.1109/cifer.1997.618940
[Accessed 29 February 2020].
[13] Tao Guo and Gui-Yang Li, "Neural data mining for credit card fraud
detection", 2008 International Conference on Machine Learning and
Cybernetics, 2008. Available: 10.1109/icmlc.2008.4621035 [Accessed
29 February 2020].
[14] M. Syeda, Yan-Qing Zhang and Yi Pan, "Parallel granular neural
networks for fast credit card fraud detection", 2002 IEEE World
Congress on Computational Intelligence. 2002 IEEE International
Conference on Fuzzy Systems. FUZZ-IEEE'02. Proceedings (Cat.
No.02CH37291). Available: 10.1109/fuzz.2002.1005055 [Accessed
29 February 2020].
[15] P. R and S. L, "Credit card fraud detection using neural network",
International Journal of Soft Computing and Engineering, vol. 1, pp.
32-38, 2011. [Accessed 29 February 2020].
[16] A. Shen, R. Tong and Y. Deng, "Application of Classification Models
on Credit Card Fraud Detection", 2007 International Conference on
Service Systems and Service Management, 2007. Available:
10.1109/icsssm.2007.4280163 [Accessed 29 February 2020].
[17] Y. Sahin and E. Duman, "Detecting credit card fraud by ANN and
logistic regression", 2011 International Symposium on Innovations in
Intelligent Systems and Applications, 2011. Available:
10.1109/inista.2011.5946108 [Accessed 29 February 2020].
[18] S. Panigrahi, A. Kundu, S. Sural and A. Majumdar, "Credit card fraud
detection: A fusion approach using Dempster–Shafer theory and
Bayesian learning", Information Fusion, vol. 10, no. 4, pp. 354-363,
2009. Available: 10.1016/j.inffus.2008.04.001 [Accessed 29 February
2020].
[19] Y. Sahin and E. Duman, "Detecting credit card fraud by decision trees
and support vector machines", in the International MultiConference of
Engineers and Computer Scientists (IMECS), Hong Kong, 2011.
[20] A. Srivastava, A. Kundu, S. Sural and A. Majumdar, "Credit Card
Fraud Detection Using Hidden Markov Model", IEEE Transactions on
Dependable and Secure Computing, vol. 5, no. 1, pp. 37-48, 2008.
Available: 10.1109/tdsc.2007.70228 [Accessed 29 February 2020].
[21] D. Shailesh S. and G. R. Bamnote, "Credit Card Fraud Detection Using
Hidden Markov Model", International Journal of Advanced Research
in Computer Science, vol. 3, no. 3, pp. 816-820, 2012. [Accessed 29
February 2020].
[22] A. Khan, T. Singh and A. Sinhal, "Implement credit card fraudulent
detection system using observation probabilistic in hidden Markov
model", 2012 Nirma University International Conference on
Engineering (NUiCONE), 2012. Available:
10.1109/nuicone.2012.6493206 [Accessed 29 February 2020].
[23] A. Prakash and C. Chandrasekar, "An Optimized Multiple Semi-
Hidden Markov Model for Credit Card Fraud Detection", Indian
Journal of Science and Technology, vol. 8, no. 2, p. 165, 2015.
Available: 10.17485/ijst/2015/v8i2/58081 [Accessed 29 February
2020].
[24] Bhusari and S. Patil, "Application of Hidden Markov Model in Credit
Card Fraud Detection", International Journal of Distributed and
Parallel systems, vol. 2, no. 6, pp. 203-211, 2011. Available:
10.5121/ijdps.2011.2618 [Accessed 29 February 2020].
[25] I. Avinash and R. C. Thool, "Credit card fraud detection using Hidden
Markov Model and its performance", International Journal of
Advanced Research In Computer Science and Software Engineering
(IJARCSSE), vol. 3, no. 6, 2020. [Accessed 29 February 2020].
[26] D. Sánchez, M. Vila, L. Cerda and J. Serrano, "Association rules
applied to credit card fraud detection", Expert Systems with
Applications, vol. 36, no. 2, pp. 3630-3640, 2009. Available:
10.1016/j.eswa.2008.02.001 [Accessed 29 February 2020].
[27] J. Quah and M. Sriganesh, "Real-time credit card fraud detection using
computational intelligence", Expert Systems with Applications, vol. 35,
no. 4, pp. 1721-1732, 2008. Available: 10.1016/j.eswa.2007.08.093
[Accessed 29 February 2020].
[28] V. Ratnam Ganji and S. Naga Prasad Mannem, "Credit card fraud
detection using anti-k nearest neighbor algorithm", International
Journal on Computer Science and Engineering (IJCSE), vol. 4, no. 6,
pp. 1035-1039, 2012. [Accessed 29 February 2020].
[29] Chuang-Cheng Chiu and Chieh-Yuan Tsai, "A Web services-based
collaborative scheme for credit card fraud detection", IEEE
International Conference on e-Technology, e-Commerce and e-
Service, 2004. EEE '04. 2004, 2004. Available:
10.1109/eee.2004.1287306 [Accessed 29 February 2020].
[30] "1.5. Stochastic Gradient Descent — scikit-learn 0.22.2
documentation", Scikit-learn.org, 2020. [Online]. Available:
https://scikit-learn.org/stable/modules/sgd.html. [Accessed: 04- Apr-
2020].
[31] "Classification Algorithms - Random Forest - Tutorialspoint",
Tutorialspoint.com, 2020. [Online]. Available:
https://www.tutorialspoint.com/machine_learning_with_python/mach
ine_learning_with_python_classification_algorithms_random_forest.
htm. [Accessed: 04- Apr- 2020].
[32] "What is the algorithm of J48 decision tree for classification ?",
ResearchGate, 2020. [Online]. Available:
https://www.researchgate.net/post/What_is_the_algorithm_of_J48_de
cision_tree_for_classification. [Accessed: 04- Apr- 2020].
[33] "Machine Learning - K-Nearest Neighbors (KNN) algorithm - Instance
based learning [Gerardnico - The Data Blog]", Gerardnico.com, 2020.
[Online]. Available: https://gerardnico.com/data_mining/knn.
[Accessed: 04- Apr- 2020].
[34] "Index of /commugrate/data/credit_card", Cs.purdue.edu, 2020.
[Online]. Available:
https://www.cs.purdue.edu/commugrate/data/credit_card/. [Accessed:
04- Apr- 2020].
[35] M. Nur-E-Arefin and M. Sultan Mahmud, "A Comparative Study of
Machine Learning Classifiers for Credit Card Fraud Detection", IJITIS,
vol. 3, no. 1, pp. 395-406, 2020. Available:
10.15157/IJITIS.2020.3.1.395-406 [Accessed 18 April 2020].
[36] A. Hammouri, M. Hammad, M. Alnabhan and F. Alsarayrah,
"Software Bug Prediction using Machine Learning Approach",
International Journal of Advanced Computer Science and
Applications, vol. 9, no. 2, 2018. Available:
10.14569/ijacsa.2018.090212 [Accessed 18 April 2020].
[37] K. Seeja and M. Zareapoor, "FraudMiner: A Novel Credit Card Fraud
Detection Model Based on Frequent Itemset Mining", The Scientific
World Journal, vol. 2014, pp. 1-10, 2014. Available:
10.1155/2014/252797 [Accessed 18 April 2020].
2020 International Conference on Innovation and Intelligence for Informatics, Computing and Technologies
(3ICT)
