Credit Card Fraud Identification: A Complete Review

University
*** Semester
Credit Card Fraud Identification
complete Review: Past and Present
Student Name:
Register Number:
Submission Date:

Credit Card Fraud Identification, a complete Review: Past and Present
Abstract
997458
The conducted research work aims on reviewing and examining thirty journals of different
publication to understand the system which detects the credit card fraud. This research's main
purpose refers idenitfying the system's frauds and to search for the best algorithm which
decreases the credit card frauds. This is regarded as a challenging task, because of the existance
of many tools, techniques or methods for the fraud detection. For each data transaction fraud
detection, distinct algorithms are utilized, for example the clustering methods, machine learning
algorithms and so on. The completion of this analysis determines that the Random forest
algorithm is the nest algorithm or the solution which could detect the credit card frauds. This
specific algorithm is selected from the review of 30 journals, which meets parameters of cost
effectiveness and system efficiency. It is extremely important to implement an algorithm which is
powerful enough, as the algorithms are the main component which detects the credit card frauds.
Moreover, it ensures to decline the number of frauds and decreases the harms of theft and frauds.
Thus, several algorithms and methods have been reviewed, followed by visualization, system
classification, system evaluation and system validation. The system classification process
concentrates on the data such as device number, Data Type, field of study, fraud detection
technique etc. Then, the system evaluation is conducted depending on the factors such as data
collection from a transaction set, the algorithms for fraud detection, analyzing of data, data
storing, data classification based on a behavioral pattern, control layers, decision making fora
transaction with respect to fraud, and proving customer notification. It is determined that the
utilization of productive and result oriented algorithms and procedures, one can meet the
objectives which have zero tolerance for the credit card frauds. The statistics and application of
algorithms are used for detecting the credit card frauds for both online and off line transactions
of the end users.
Keywords: Supervised & Unsupervised Models, Credit card cheating, Shams, Algorithms,
Identify Fraud, Data Mining.
1

1. Introduction
In today's modern world of advanced technology, it is vital to be aware of affects related to fraud
and theft. Specially related to financial and money related frauds. Lot of research work is being
conducted and Technology invented to reduce the amount of frauds associated with credit cards.
The main concern of banks, companies, firms, individuals, end user is credit card theft and how
this can be prevented all together. These theft and frauds can be of different types. We shall
discuss and try to understand this issue and the problem faced by the industry.
Normal framework, system servers and networks are vulnerable to outside potential threats and
security attacks. This can be due to unreliable and not maintained infrastructure, no anti-virus
precautions, no reliable malwares, low security, and data compromise. Understanding the past
history and past cases, records related to frauds and thefts, we can predict and can become
proactive against any possible attack on the credit card data. By analysing the past data we can
understand behaviorof fraudster’s and hackers. A wide spectrum of data classification and
discipline is required to maintain a proper anti-theft policy and strong safety features. The role of
the system administrator and supervisor becomes vital and important for maintaining the
standard of system/ Networkswhich are related to credit card data and information.
Classify the event as a fraud and a threat, we shall make use of the supervise model to make
absolute certain and for classification of that event which was studied by Bolton and Hand in
2002. Few measures are provided for identifying and for classifying the events. Events which are
abnormal, uncommon can be detected by the supervised methods and procedures. To detect the
various types of frauds, a fraud model can be utilized. As per the symptoms of an event which is
related historical fraudulent cases, the abnormality is classified. Further investigation should be
done on such cases and this will not prove that these current events are suspicious and
fraudulent. This is just a start for more data gathering, information, collection of evidence and
further investigation study to determine the truth of the case. (Nisbet, Miner, Yale, Elder &
Peterson, 2018; Kültür, Y., & Çağlayan, M. U., 2017; Bahnsen et al., 2017; Zhao et al.,
2016).Active learning strategies can be beneficial for fraud detection (Carcillo, F., 2018).
Online shopping has become the new craze in today’s world with rapid increase and penetration
2

of internet in the whole world. There is a demand for e-commerce and thus this leads to multi
users on the web doing many transactions by using their cards, net banking, credit cards, e-
wallets etc. The most used out of these all transaction modes is the Credit card. More the credit
card is used on e-commerce platforms, the more the chances of theft and stolen data. These all
ultimately leads to financial thefts and losses. Now the authors of these papers have come up
with an innovative method to detect such frauds. Understanding and analysing the behavioral
patterns and transaction data of the credit card user is taken and processed into an algorithm to
get a certain pattern of behavior. This specific type of data of the cardholder is then used to
identify the online fraud when it detects a new fraud. This new thought of conceptual drift is
used as a feedback mechanism for detecting the process and ultimately solving the fraud.
(Jiang, Song, Liu, Zheng & Luan, 2018).
The ultimate aim of this project and study is to come up with the best algorithm that will detect
the frauds in credit card and the various types of theft/ frauds that can occur and how to detect
them. They have been able to identify a few proven methods for solving this enigma. Data
mining, Machine learning, Forward-Backward Greedy algorithm, Artificial Neural Network,
Decision Tree etc are some of the technics and procedures that will be considered.
Understanding the process of evaluating and picking up the right method in the
detection of credit card fraud by validating the process and considering various tested methods
based on research paper analysis based on the outlined data information, Techniques and
components. Pinpointing and identifying the future developments based on the past historical
data and comparing it with present collected data.
Evaluation, study and analysis will be carried out for identifying the credit card hacks, thefts and
frauds on different platforms and networks with the aid of custom made algorithms. Assistance
of the two types of classification that was defined before i.e. Supervised and Unsupervised
Systems will be taken to identify the perfect direction in pinpointing these frauds on any of the
platforms and networks being used.
Lot of hard work and efforts have been put in by these authors to solve and come up with an
answer for this issue and how to overcome and benefit by using their researched technic. The
3

parameters that we shall use when using these technics will be simply be the accuracy,
effectiveness, and cost.
Now, the process will include the following steps,
Firstly we shall determine the process of detecting and identifying a fraud. There will be a
discussion on this matter and the best possible method will be selected.
Second step will be to review and study research papers which have identified previous cases
and how they were used for thefts and frauds.
Third, we shall come up with the technics and the methods to deal with the frauds. Selection of
the best possible technic is analyzed and chosen out of the lot. Subjecting the system to various
tests and identifying the areas also will be taken into consideration. The best 12 algorithms will
be selected and reviewed as to how they will be used and how they will prevent future frauds in
the system.
The Best working Algorithm out of the 12 algorithms shall be taken for identifying the credit
card frauds and thefts and which will be part of the final stage of our project.
With the ever increasing population of online users and the numerous transactions that take place
online by individuals using their credit cards, it becomes more than ever to understand and make
a formal classification for frauds and thefts related to credit cards. Development of various fraud
detection methods and procedures are being developed and researched by various groups with
use of individual data and information, evaluating the same and their credit card usage history.
(Jiang, Song, Liu, Zheng &Luan, 2018; Jiang et al., 2018; Bahnsen et al., 2016; Ganji &
Mannem, 2016).
2. Literature Review
We shall now understand and share our views on the various algorithms and different types of
procedures and technics in Credit card fraud identification. The 12 chosen algorithms as per
previous discussion will be considered and analyzed as to how best they can be used against the
credit card frauds.
4

2.1 Fraud-BNC Algorithm
A specific Bayesian Network Classifier (BNC) algorithm based on the principle ofFraud-
BNCmethod and was designed by de Sá, A. G., Pereira, A. C., & Pappa, G. L. (2018) which shall
use data and information from all the credit card frauds, history and data in the real world issues.
The idea behind this is to better the data classification for identifying thedeceit and malpractice.
The usage and support of the Bayesian algorithm which is customized is taken. The main
problem with this method is that the Data categorization was not used and implemented properly
for identifying the frauds and thefts. Applying and using this for Data classification and
unbalanced data and by using the fraud detection techniqueswhich again are used with the help
of customized BNC algorithm, economic metrics’ efficiency, classification of metrics and
elements of the Fraud detection algorithm can solve this issue. To help this classification of
imbalanced data we shall use the better featuresof the given system. This shall decrease 72
percent of fraud tractions once it is implemented in the system. This method will be used to
identifying the Fraud detection and problems which can lead to financial damages to the
company. The system will keep the information of all the deals and transactions that includes the
end users, companies and organizations. (Bahnsen et al., 2016; Save et al., 2017).Transaction’s
authorization is granted only if the permittedthreshold value is passed or else there will be no
Transaction(de Sá, Pereira & Pappa, 2018).
2.2 Random Forest Algorithm
The root cause of Card frauds is because of all the information that is declared by the user which
includes the card name and personal information. This leads to hacks and card frauds from the
system. (Niveditha, G., Abarna, K., & Akshaya, G. (2019). By correctly identifying, checking
and having a control on these will minimize these issues. A trained and programmed model based
on standard attributes is used for monitoring the transactions by users and the behavior and
actions of this user is tested by this model. If this model senses some irregularity as per its
parameter than it will reject the transaction. Synthetic Minority Over-sampling Technique or
“SMOTE” can be used for this purpose and which is based on a statistical method. (Fiore et al.,
2017; Jurgovsky et al., 2018; Dal et al., 2018). Development of multiple trees and integrating all
these data and information on the entire tree gives good results, whose base is based on the
5

random tree principle. Thus this gives very good and accurate results i.e., about 98.6
percent(Niveditha, Abarna & Akshaya, 2019).
2.3 Machine Learning Algorithms
A lot of scholarly papers and research analysis have encouraged the use of machine
learning algorithm (Jurgovsky, J., Granitzer, M., Ziegler, K., Calabretto, S., Portier, P., He-
Guelton, L., & Cael). There is an increasing substantial challenge for the financial organizations
and for the service providers as there is increase in online transactions by users in way of
electronic payments and by ec0mmerce portals and platforms. (Fan, Han & Liu, 2017; Fahmi,
Hamdy & Nagati,2016; Duchi, Hazan & Singer, 2011; Fercoq, & Richtárik, 2018). A slow and
somewhat time lag procedure which is caused due to the slowly creeping in of the above method
way into the business applications because of its demand and accuracy but which overall results
into detection of the various theft and frauds.The speed to detectthe frauds on o ine transactionsffl
has to be increased as per the rapid increase in technology and online users. For the offline
transaction frauds, machine learning has proved to be more effective and penetrative with the
help of the method as discussed. Long Short-Term Memory (LSTM) is one such method that is
used for this process and it has proved to have a very good detection capability. (Jurgovsky et al.,
2018; Aleskerov, Freisleben & Rao, 2016).
2.4 ANN
Information from the day-to-day transaction will be used in the identification and understating
the characteristic features of a normal or fraudulent transaction and this is also taken in the
reporting and data from the historical transactions cases. (Choi, D., & Lee, K. 2018; Zareapoor &
Shamsolmoali, 2015).In present world, more importance is to incorporate an effective system to
detect the credit card defects and frauds as it is a common thing and the usage is rampant and
very frequent. (M. Simi. (2009)). Random Forest, ANN (Artificial Neural Network) and SVM
(Support Vector Machine) are some of the most common machine learning supervised technics
and algorithms aids in detecting and determining this detection of online credit frauds and hacks.
It has very good detection and accuracy level when compared to the other algorithms such as
ANN and SVM (Rajeshwari & Babu, 2017; M, 2009; Gómez et al., 2018;Carcillo et al.,2018).
6

2.5 Data Mining Algorithms
There can be lot of complications and confusions in large transactions and its related
frauds.(Zanin, M., Romance, M., Moral, S., & Criado, R. (2018). Here the importance of data
mining algorithms comes into picture. This problem of absence of metrics synthesizing to initiate
the global structure for data mining algorithms can be removed by the advanced system proposal,
which uses the proper and accurate integration of network metrics. There is a marked
improvement in the classification.Banks are the most beneficial from this technique which
arranges and manages the complicated organizational projects.16% to 12% drop in error rate
proves its power and detection success. For the overall success and improvement of data
detection frauds, combined use of neural network and data mining should be considered and
utilized. Special mention should be made of the super and more reliable method in use of neural
network for its high accuracy and detection in credit card cases. (Fu et al., 2017; Zanin et al.,
2018; Yee, Sagadevan & Malim, 2018).
2.6 Anti-k nearest Neighbor Algorithm
CardCom is the one utilized todetect the fraud as per the research paper (Robinson, W.
N., & Aria, A. (2018)). The card’s sequence, unique approach with numbering method is used for
the fraud detection in various organizations. If the data usage and the data is modified than the
validate transaction, recall and result all will change and be different. Raw and unused Data plays
a very important role in detection resultsand also for sorting and evaluating these raw data
The machine learning and classifiers’ implementation will reduce the drawbacks’ in CardCom’s
limitations.( Bahnsen et al., 2016; Ganji & Mannem, 2016).
2.7 Decision Tree Algorithm
Various opportunities come along for detecting credit card frauds and thefts as perCarta,
S., Fenu, G., Recupero, D. R., & Saia, R. (2019). This paper gives their view point that there will
be viruses, bugs and Credit card thefts and frauds even if the system and the network has all the
malwares, anti-virus software’s and programs. (Vaughan, G., 2018). Use of the (Prudential
Multiple Consensus) PMC model is taken to apply this research paper to the e-commerce
business. With the ever increasing presence of individuals on the net and thus a direct increase in
the transactions over e-commerce trades and buying will lead to more credit card related issues
7