Volume 7, Issue 5, May – 2022 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Credit Card Duplicity Spotting on Gaining the

Knowledge on Machine Learning
T. Hemanth 1, K. Nikhil 2, *, K. Dheeraj 3, Srikanthyadav. M4
1,2,3
UG student, Department of Infomation Technology,
4
Department of Infomation Technology,
Vignan’s Foundation for Science, Technology & Research, Guntur AP-522213, India

Abstract:- With the help of technologies like artificial B. Manual Credit Card Imprints
intelligence (AI), machine learning, big data, blockchain, The magnetic strip on a credit card is used by fraudsters
cloud computing, and IoT the technological revolution is in this sort of scam. There will be falsified transactions in the
speeding (IoT). There has been a dramatic increase in the future because of this knowledge.
number of cyber-attacks and criminal activities as a result
of the widespread use of ever-improving internet C. Original Card Not Exist
technologies. Fraudulent use of credit cards is a major When a card is used without the cardholder actually
concern for the banking sector across the world. Credit having it in their hands, the fraudster gets access to the
card fraud is growing at an alarming rate and has become account number and expiration date of the card.
a major concern, especially as the amount of financial
transactions utilising credit cards grows. Here, we've D. Mail Non-Receipt Fraud
looked at some credit card fraud detection methods that Every time a consumer applies for a new credit card, the
can help protect against a variety of scams. The research process takes a while. Because of this, fraudsters utilise
problems were also discussed and analysed. For the intercepted delivery to their advantage, changing the user's
purpose of detecting credit card fraud, we've deployed six identity to their own and then making transactions, a practise
widely-accepted machine learning approaches. A known as Never Received Issue Fraud.
confusion matrix is created for each machine learning
approach so that the algorithm's performance may be E. Counterfeit Card Fraud
evaluated. Accuracy, precision, recall, specificity, All the properties of a genuine magnetic swipe card may
misclassification and F1 score are used to evaluate its be found in this replica card. Skimming is a method that may
efficacy. Machine learning approaches can be useful in be used to accomplish this. This fictitious card may be used
detecting credit card fraud, according to the results. For to make purchases because it is completely working.
fraud detection, we propose utilising different machine
learning algorithms, even though the findings F. Off track and Stolen Card Fraud
demonstrate that each algorithm has a high degree of Due to some unknown circumstances, the card holder
precision and recall. had their card stolen. If a fraudster manages to obtain the card,
he or she may then use it to make purchases. Because internet
Keywords:- Credit Card Fraud, Fraud Detection by Machine transactions require a pin number, it's harder, but it can still
Learning, Machine Learning Techniques. be done.

I. INTRODUCTION G. False Merchant Sites

Phishing is a sort of fraud that resembles this type of
Credit cards are increasingly being used to pay bills and scam. When a fraudster creates a false website, it has the
conduct online purchases. As a result of this shift in payment appearance of being authentic. Customers will be enticed by
methods, funds are now transmitted via digital means. With the appealing designs and incentives, such as significant
these cards, cashless systems have been worsened and cash discounts, buy one, get one free, etc. Once a transaction has
credit has been relieved at the same time. At the same time, been completed, the cardholder's information is gathered and
the usage of credit cards is on the rise, and this has led to an stored. This might be utilised in the future to conduct
increase in credit card fraud. One of the most common types fraudulent transactions.
of credit card fraud is perpetrated by people who don't intend
to pay back the money they've borrowed. A variety of credit It is possible to identify credit card thefts using historical
card-related scams are described in detail in this section. data by analysing the varied purchase habits of a certain
customer. Banks and other credit card issuers can benefit
A. Application Fraud from this data analysis by reducing the amount of money they
The fraudster creates a phoney user account in order to lose to credit card fraud. When comparing historical data with
get access to sensitive information like login and password, current purchase habits, a statistical model is needed to
and then manipulates the application framework using this identify fraudulent tendencies and alert banks to suspicious
account. To make things worse, he steals customer service activity. This enables banks to quickly identify and eradicate
records. CC frauds by rejecting suspicious transactions, hence
reducing the risk of fraud.

IJISRT22MAY2003 www.ijisrt.com 1777

Volume 7, Issue 5, May – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Various machine learning approaches are employed in In this study, researchers used a Hidden Markov Model
this work for the goal of identifying these scams, and they are (HMM) to identify credit card thefts with brief false alarms.
used in this study as well. The accuracy of fraud detection is It is possible to detect fraudulent credit card transactions
directly tied to the parameters of the evaluation. The values using neural networks and support vector machines,
of these factors make it easier to determine whether or not the according to the authors [10]. An RBF neural network with
transaction is legitimate. The major goal of this study is to three layers of feed-forward is constructed.
evaluate the effectiveness of several machine learning
algorithms for detecting credit card fraud. Genetic algorithms (GA) were employed to choose
support vectors in a Binary Support Vector System (BSVS)
II. RELATED WORK that was shown in this study [11]. A real negative rate was
first determined by using the Self-Organizing Map (SOM)
Various academics have utilised a variety of methods to method, and then the data was trained using BSVS. An ANN
identify credit card fraud. Some of the noteworthy credit card (Artificial Neural Network) and decision tree hybrid was
fraud detection work was reviewed in this section. [1] There employed in this study [12].
are a number of artificial intelligence systems that use the
physical unclonable functions [1] to facilitate and protect Testing and implementation were carried out using a
electronic transactions between distinct entities. two-phased method. The dataset was generated in the first
step using the classification outputs of decision trees and
Various classifiers' performance is studied in this multilayer perceptrons. Feed this dataset into a multilayer
research [2] and a summary of the results is provided. [3] perceptron for data categorization in the second phase. Phase
employed an ANN-GA hybrid model for credit card fraud 2: This model has a low percentage of false positives, which
detection in this article. ANNs are artificial neural networks, means it's dependable. A four-stage Bayesian network for
and GAs are genetic algorithms. To classify transactions, a fraud detection was created by the authors of this research
neural network and a genetic algorithm have been deployed, [13]. As a result, they concluded that their suggested method
respectively. would be excessively slow in comparison to other popular
algorithms like K-nearest neighbour, neural networks, and
In this research, the author [4] describes how a neural regression.
network may be used to mine data to detect credit card fraud.
Auto-associative architectures are employed in three levels to The authors of this research [14] used a two-phase
get the desired results in this project. They trained and tested hybrid approach that incorporates neural networks and fuzzy
the system on a set of synthetic data. With this solution, they clustering. The c-means clustering algorithm was proposed in
were able to obtain extraordinarily high fraud detection rates. phase one. In the second step, they feed suspicious
transactions into the neural network to determine if it was
The authors of an article [5] suggested a model for real- fraudulent or not.
time fraud detection based on bidirectional neural networks.
A model based on an Artificial Immune System was In study [15], the authors employed Bayesian networks
presented by the authors [6] of this research for the and ANN, two machine learning approaches, to detect credit
identification of credit card theft in the online environment. card fraud. ANNs were also discussed as a way to speed up
To ensure that they all performed at the same level, a logistic Bayesian networks after only a minimal amount of training.
regression model and three different methods were utilised. Fuzzy logic and neural networks were used to create a system
for fraud detection in this research [16]. They found that ANN
The authors of this study [7] describe a model for was 33 percent more accurate than fuzzy logic in terms of
detecting credit card fraud based on the principles of a genetic precision. Fuzzy logic was utilised to assign a membership
algorithm. In this method, a two-phase synthesising algorithm characteristic to each piece of data already in the system. The
was devised. Data generation is carried out in the first step neural network was utilised for results validation.
before algorithms are applied to detect fraudulent transactions
in the second phase. III. METHODOLOGY

The authors of this study [8] discuss a genetically We took these procedures to deploy the machine
programmed fuzzy system for detecting credit card fraud. learning algorithms outlined above. On Kaggle [17], a
Using this method, data from actual house insurance claims standard data set is used to evaluate the algorithms. Python
and credit card transactions may be utilised to generate has been used to provide a test environment in which various
categorization criteria. machine learning approaches may be evaluated.

IJISRT22MAY2003 www.ijisrt.com 1778

Volume 7, Issue 5, May – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

A. Data Set Description parameters are used to calculate accuracy, recall, and
In order to obtain this data, you may go to: https://mlg- precision, among other things.
ulb/creditcardfraud/home. Due to difficulties with
confidentiality, this dataset does not include any background
information on the 28 features out of the 30 features. Thus,
principal component analysis generates the values of these
attributes. Principal components such as 'Amount' and 'Time'
have not been turned into features. When fraud occurs, the
class is set to 1 and when it does not occur, the class is set to
0.

B. Performance Evaluation Parameters

A confusion matrix is a common tool for assessing the
effectiveness of a classification statistical model. The target
class number N is used to create a confusion matrix. The N x
N matrix provides us with a full view of our model's
performance and the kinds of errors it is committing.
IV. RESULTS AND DISCUSSION
According to actual and expected positive and negative
The outcomes of the machine learning approaches
class values (see fig. 1), there are four parameters shown. TP
outlined above are examined and demonstrated in this part.
and TN refer to the values when the model accurately predicts
Dataset anomalies and how to use them for training and
the positive or negative class. For example, when a model
predictions something wrongly in a positive or negative validation of datasets must be understood for performance
category, it is known as a False Positive (FP). analysis. Overall, there are 29,415 normal instances and 592
bogus ones in the dataset, with a total of 294807. A parameter
value for the confusion matrix of each approach is calculated
Any machine learning technique's performance is
evaluated using the confusion matrix parameters. Based on in table 1 using normal and fraudulent scenarios as examples.
assessing the many sorts of values, such as true positives, The estimated values of the evaluation parameters for each
false negative and the like in the single confusion matrix, approach are shown in Table 1. The amount of precision
these parameters have been developed. Confidence matrix achieved by each approach is almost the same.

Table.1. Performance analysis of ML approaches

Technique Accuracy Precision Recall Specificity Fl Score
Decision Tree 0.9991 0.9995 0.9995 0.7484 0.9995
Isolation Forest 0.9995 0.9996 0.9999 0.7707 0.9997
K-NN 0.9995 0.9996 0.9999 0.8846 0.9997
Logistic Regression 0.9992 0.9994 0.9998 0.6730 0.9996
Random Forest 0.9993 0.9996 0.9997 0.7799 0.9997
Support Vector Machine 0.9995 0.9996 0.9999 0.7692 0.9997

IJISRT22MAY2003 www.ijisrt.com 1779

Volume 7, Issue 5, May – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
V. CONCLUSION [9]. Bhusari, V., & Patil, S. (2011). Application of hidden
markov model in credit card fraud detec-tion.
For the purpose of detecting credit card fraud, we used International Journal of Distributed and Parallel
six different machine learning approaches, which we tested in Systems, 2(6), 203.
this article. A close examination of the various methods [10]. Ghosh, S., & Reilly, D. L. (1994, January). Credit card
reveals that not all of the procedures yielded the same results fraud detection with a neural-network. In System
in all circumstances. The results of fraud detection algorithms Sciences, 1994. Proceedings of the Twenty-Seventh
are also dependent on the type of dataset that's being Hawaii International Conference on (Vol. 3, pp. 621-
analysed. Techniques that yield excellent accuracy but 630). IEEE.
require a lot of time and money to learn are available. When [11]. Chen, R. C., Chen, T. S., & Lin, C. C. (2006). A new
used to huge datasets, certain strategies produce subpar binary support vector system for increasing detection
outcomes when applied to small data sets only. Some rate of credit card fraud. International Journal of Pattern
strategies perform better with pre-processed and sampled Recognition and Artificial Intelligence, 20(02), 227-
data, whereas others perform better with raw, unprocessed 239.
data. Another thing to keep in mind when dealing with [12]. Keskar, V. (2020). Comparing different models for
anomaly identification is that the outlier class of modelling credit card fraud detection. Journal of Critical Reviews,
might be useless or even harmful. Focus must be placed on 7(2), 981-986.
the structure and distribution of typical data. Detecting credit [13]. Ezawa, K. J., & Norton, S. W. (1996). Constructing
card fraud while it is still in transit requires a system that can Bayesian networks to predict uncollectible
detect fraudulent activity and identify it precisely, with as few telecommunications accounts. IEEE Expert, 11(5), 45-
classification errors as possible. 51.
[14]. Behera, T. K., & Panigrahi, S. (2015, May). Credit card
REFERENCES fraud detection: a hybrid approach using fuzzy
clustering & neural network. In 2015 Second
[1]. Fragkos, G., Minwalla, C., Plusquellic, J., & International Conference on Advances in Computing
Tsiropoulou, E. E. (2020). Artificially Intelligent and Communication Engineering (pp. 494-499). IEEE.
Electronic Money. IEEE Consumer Electronics [15]. Maes, S., Tuyls, K., Vanschoenwinkel, B., &
Magazine, 10(4), 81-89. Manderick, B. (2002, January). Credit card fraud
[2]. Awoyemi, J. O., Adetunmbi, A. O., & Oluwadare, S. A. detection using Bayesian and neural networks. In
(2017, October). Credit card fraud detection using Proceedings of the 1st international naiso congress on
machine learning techniques: A comparative analysis. neuro fuzzy technologies (pp. 261-270).
In 2017 International Conference on Computing [16]. Razooqi, T., Khurana, P., Raahemifar, K., & Abhari, A.
Networking and Informatics (ICCNI) (pp. 1-9). IEEE. (2016, April). Credit card fraud detection using fuzzy
[3]. Patidar, R., & Sharma, L. (2011). Credit card fraud logic and neural network. In Proceedings of the 19th
detection using neural network. International Journal of Communications & Networking Symposium (pp. 1-5).
Soft Computing and Engineering (IJSCE), 1(32-38). [17]. Kaggle: Your Machine Learning and Data Science
[4]. G Aleskerov, E., Freisleben, B., & Rao, B. (1997, Community. Kaggle.com. (2021). Retrieved 8 May
March). Cardwatch: A neural network based da-tabase 2021, from https://www.kaggle.com/.
mining system for credit card fraud detection. In
Proceedings of the IEEE/IAFE 1997 computational
intelligence for financial engineering (CIFEr) (pp. 220-
226). IEEE.
[5]. Krenker, A., Volk, M., Sedlar, U., Bešter, J., & Kos, A.
(2009). Bidirectional Artificial Neural Networks for
Mobile‐Phone Fraud Detection. Etri Journal, 31(1), 92-
94.
[6]. Brabazon, A., Cahill, J., Keenan, P., & Walsh, D. (2010,
July). Identifying online credit card fraud using artificial
immune systems. In IEEE Congress on Evolutionary
Computation (pp. 1-7). IEEE.
[7]. RamaKalyani, K., & UmaDevi, D. (2012). Fraud
detection of credit card payment system by genetic
algorithm. International Journal of Scientific &
Engineering Research, 3(7), 1-6.
[8]. Bentley, P. J., Kim, J. W., Jung, G. H., & Choi, J. U.
(2000). Fuzzy darwinian detection of credit card fraud.
In Proceedings of the Korea Information Processing
Society Conference (pp. 277-280). Korea Information
Processing Society.

IJISRT22MAY2003 www.ijisrt.com 1780