INTRODUCTION 

Each application that is running today requires large size of 
databases such that huge amount of data that is being 
generated within them can be stored in them. There is a need 
to generate new and highly efcient tools which can manage 
such huge databases and the data present within them. For 
facilitating the large number of users as well as their 
applications several techniques have been proposed by 
researchers. Amongst such techniques is the data mining and 
knowledge discovery approach that is used in order to 
manage all such data. Data mining is known as the 
automated approach of identifying interesting patterns from 
huge databases [1]. This technique helps in generating 
descriptive, understandable and predictive models from the 
existing data within the databases. 

The main reason behind the development of this work is to 
store data which is generated each day due to all the 
applications has been utilized by humans. In the computer 
databases, this data is stored so that it becomes easy to 
access data whenever it is required. The similar objects are 
separated from each other amongst different groups this 
process is known as clustering. Cluster is dened as the group 
in which similar objects are placed within the same group [2]. 
Similarly, objects having different properties and are 
dissimilar with each other are placed in separate cluster.  The 
representation of data using less clusters leads to loss of 
essential information. With the help of this approach, it 
becomes easy to simply data in controlled manner. In the 
databases large amount of data is present which is stored 
from long time. The extraction of this information is very 
important in order to utilize it in useful manner [3]. The process 
of data mining extracts the useful information with quality and 
also provides effective sharing. The discovery of predictive 
intelligence has been done with the help of uncovering 
patterns and relationships present within structured as well as 
unstructured data in case of data mining and text analytics 
[4]. According to researchers, it is required to have deep 
knowledge or creative skills so that models can be generated 
within predictive analytics. Hence, this approach is developed 
with the help of some basic functions. The use of credit card is 
common nowadays due to change in technology from 
moderate to modern. Therefore, it is necessary to develop a 
model for the detection of frauds in credit cards in all regions. 
Fraud in credit cards is done by any unauthorized account 
activity of a person [5].  Thus, it is necessary to take some 
appropriate steps by which increasing frauds and threats can 
be minimized timely. For the protection of the clients various 

methods are facilitated that provide the risk management that 
eliminates the misuses of credit cards misuse. The credit card 
fraud took place when user as well as issuer has no 
knowledge that other use tries to use card. Hence, this is 
known as frauds in which one card is used by another. The 
person who tries to use card of another person is not having 
good intention as that person want to withdraw all money from 
user account. The unauthorized user is identied with the help 
of fraud detection system, if one could commit the fraud. In 
order to minimize all these frauds and to attract criminals 
various fraud detection methods has been developed so far 
using various strategies [6]. The development of new 
mechanism for fraud detection is becomes difcult due to the 
limitations of exchanging the ideas of fraud detection. 

Literature Review 

Kuldeep Randhawa, et.al (2017) presented the use of machine 
learning algorithm has been utilized for the detection of the 
fraud in the credit card by developing a system [7]. The 
standard models was utilized initially after which the use of 
hybrid methods is followed in which AdaBoost is present. The 
efciency of the model has been evaluated with the help of 
credit card dataset which is available everywhere. In the data 
samples noise is embedded in order to evaluate the 
robustness of algorithms. On the basis of performed 
experiments, it is concluded that fraud detection methods 
provides the higher accuracy rates. With the help of this 
approach online learning models will be proposed in future to 
extent the work. Hence, it becomes easy to identify fraud cases 
quickly using online learning mechanism.

Suman Arora, et.al (2017) presented various issues are faced 
by individual while selecting appropriate fraud detection 
models as per done study [8]. The implementation of an 
effective process coefcient sum mechanism was done whole 
set of selection criteria and FDMs are introduced. For the 
analysis of various scenarios, this proposed approach has 
been utilized using decision maker's criterion. There is no 
fraud detection model till yet which satises all the needs by 
which frauds can be detected easily. They proposed a method 
in this paper which is based on the integration of different 
comparison so that fraud can be detected ranked easily. With 
the help of this proposed method, the complex multi-attributes 
decision issues can be resolved easily.  

S Md. S Askari, et.al (2017) presented the use of fuzzy logic 
which is the combination of mathematical process by which 
an ID3 decision tree system is generated in this paper. They 

MACHINE LEARNING APPROACH FOR CREDIT CARD FRAUD 
DETECTION

Original Research Paper

Geetika Student, Department of Computer Engineering, Punjabi University, Patiala

Computer Engineering 

The extraction of the useful information from the raw data is done a technique known as data mining. The 
prediction of new things from the current data has been done using the prediction analysis which is the 

application of data mining.  Classications techniques are most commonly used which are implemented for the prediction 
analysis. Hence, prediction of the credit card fraud detection is the main objective of this work. Author proposed various credit 
card fraud detection mechanisms and techniques to prevent and detect fraud timely. The fundamental of the proposed 
technique in the base paper is based on the conventional neural networks. This system drives the new values and learns from the 
previous experiences. For the detection of the credit card fraud, SVM classier is proposed in this research work using which 
input data is classied into normal and fraud transactions. Test and training sets are the two sub-parts of the input data. In terms 
of precision and recall, the normal and fraud transactions have been predicted on the basis of test and training sets. 

ABSTRACT

KEYWORDS : Fraud Detection, SVM, KNN, Naïve bayes

Dr. Gaurav Gupta*
Assistant Professor, Department of Computer Engineering, Punjabi 
University, Patiala *Corresponding Author

VOLUME-8, ISSUE-9, SEPTEMBER-2019 • PRINT ISSN No. 2277 - 8160 • DOI : 10.36106/gjra

96 X GJRA - GLOBAL JOURNAL FOR RESEARCH ANALYSIS



implemented the FuzzITree algorithm on the normalized 
training data [9]. As per performed experiments, it is 
demonstrated that all the performed transaction was 
classied correctly except for the transaction number 7. They 
performed various tests on different transactions so that 
detection rate can be determined easily in different situations. 
On the basis of achieved results, it is identied that the 
detection rate is around 89%. In future, advanced concepts of 
fuzzy will be utilized so that fraud detection can be identied 
easily.  

Luis Vergara, et.al (2017) presented in order to enhance the 
performance of credit card fraud detection systems, they 
proposed various methods which are based on signal 
processing on graphs [10].  In this paper, they proposed 
iterative amplitude adjusted Fourier transform along with 
iterative surrogate signals on graph algorithms as an 
alternative. It is necessary to do a reliable augmentation of the 
target inadequate population of frauds due to the present 
issues. These issues are labeling cost, algorithm testing, data 
condentiality and consistent modication of patterns. With 
the help of legitimate and non-legitimate transaction ratios, 
they evaluated the detection capabilities feasibility using 
receiver operating characteristic (ROC) curves and several 
key performance indicators (KPI) are utilized. 

Fabrizio Carcillo, et.al (2017) presented for the detection of 
frauds and to enhance the accuracy of fraud detection ratio 
they combine the previously proposed techniques with new 
methods in this paper [11]. They mentioned the development 
or investigation of tradeoff, in terms of fraud detection. With 
the help of corresponding machine learning techniques, they 
performed the widespread analysis. This can be done by 
transacting millions of real-world dataset which has been 
provided by an industrial system. On the basis of performed 
experiments, it is concluded that High Risk Query mechanism 
is achieved here for the enhancement using semi-supervised 
leanring.

Rajeshwari U, et.al (2016) presented for the prevention and 
detection of frauds timely, they proposed a mechanism in this 
paper. With the help of this method, alert related to fraudulent 
transactions will be sent to the card owner and credit card is 
blocked immediately [12]. The value generated for Hidden 
Markov Model outcome is the basis that decides whether the 
transaction is fraudulent or not. With the help of this system, all 
the marked fraud cases among the actual transactions can be 
minimized using this by comparing the rate of false alarms 
that are achieved. In order to minimize the false alarm rates 
and to prevent frauds they utilize the streaming analytics. 
They proposed a model and update it regularly on the basis of 
collected information of genuine card holder. 

RESEARCH METHODOLOGY 
The credit card fraud detection algorithm and data 
classication methods were evaluated by conducting a 
detailed literature study in the rst step. This was done to know 
their advantages and limitations. The limitation of the existing 
schemes is overcome by the design of proposed security 
solution that improves the structure of this method. In order to 
build a robust system, it combines all the advantages of the 
system together. For the simulation process, they utilize the 
PYTHON simulator in which the proposed solution is 
implemented using all essential input and output parameters. 
After, implementation of the proposed method they analyze 
the performance and compared this method with the existing. 
From last 10 years, the historical data of the credit cards has 
been utilized for the process of input data acquisition. In the 
historical data, all the details of transactions of every day is 
present which provides the information to user when the 
amount is withdrawn and by what time and from which 
location. By collecting the regression models with squared 

distance, they designed the proposed model during the 
process of implementation for classication. This was done for 
processing of historical data which has been utilized for the 
long term prediction. The classication of the input data into 
the normal and fraud transactions can be done using SVM 
classier in this research work. The SVM classication is also 
known as a predictive model which has been utilized for the 
text categorization. The used input here is data and output is 
the classied data given in two categories. With the help of 
SVM training algorithm, a model is implemented for the 
amount of text where each training example belongs to one of 
the two classes. The data is divided into two categories by 
constructing N-Dimensional hyperplane. On the each side of 
the hyper plane, they generated two parallel hyper planes so 
that data can be separated easily. Hence, by separating the 
hyper planes, there is increase in the distance between the two 
hyper planes. In association to the partitioning hyper plane 
f(X) divides the two classes when it passes through middle. For 
the creation of the linear classication function, there is a 
linearly separable data set. The testing of sign of function 
f(Xn) easily test and classify the new data instance, Xn when it 
determine the function.: 
Where Xn belongs to a positive class if f(Xn) > 0 
 
The generalization of the error of the classier for the larger 
distance or margin can be done in an optimal way. This 
algorithm functioning is very good in concern with high 
dimensional feature set. For the creation of the new linearly 
separable data they utilize the kernel trick in which non-
linearly separable data is transferred. SVM classication has 
been utilized for the calculation of regression analysis and to 
perform numerical calculations. This algorithm is also useful 
in ranking the elements. There are various attributes present 
on the dataset of the SVM and it performance is good as only 
specic cases can be accessed for training purpose which is 
considered as the advantage of SVM. But, it has limitation of 
speed and size which occur during the training and testing 
phase of SVM. It is not easy to choose the parameters of kernel 
function hence, consider as disadvantage of algorithm. 

Fig 1: Proposed Flowchart

EXPERIMENTAL RESULTS 
The proposed technique has been implemented in Python and 
the results have been analyzed in terms of accuracy as shown 
below.

VOLUME-8, ISSUE-9, SEPTEMBER-2019 • PRINT ISSN No. 2277 - 8160 • DOI : 10.36106/gjra

  X 97GJRA - GLOBAL JOURNAL FOR RESEARCH ANALYSIS



Figure 2: Accuracy Comparison

As shown in gure 2, the accuracy of proposed and existing 
algorithms is compared and it is analyzed that proposed 
algoirhtm performs well in terms of accuracy. 

Figure 3: Recall Analysis

As illustrated in gure 3, the recall of voting based 
classication method is compared hybrid classication 
model. The hybrid classication model is more recall as 
compared to voting based classication method for credit 
card fraud detection. 

Figure 4: Execution Time

The execution time of the voting and hybrid classication is 
compared in the gure 4. The execution time of the hybrid 
classication method is low as compared to voting based 
classication method for credit card fraud detection 

CONCLUSION 
Data mining is the technique using which essential 
information can be easily extracted from the large amount of 
data. Prediction analysis is considered as the application of 
credit card fraud detection. The prediction of future data in the 
prediction analysis has been done using current information. 
The implementation of neural networks was done in the 
existing techniques by which input data is learned that drive 
the future values. For the classication of normal and fraud 
transactions, they implemented the classication technique in 
this work. The whole data is divided into test and training sets, 
by applying the SVM classier in this work. Here, input is the 
test and training set which provides the future values. 
Experimental results, evaluated the performance of proposed 
modal as compared to existing technique in terms of accuracy. 

REFERENCES 
[1]  K. C. Tan, E. J. Teoh, Q. Yu, K. C. Goh., “A hybrid evolutionary algorithm for 

attribute selection in data mining”. Expert system with applications 2008, 
Elsevier 

[2]  Pardeep Kumar, Nitin, Vivek Kumar Sehgal, Durg Singh Chauhan. “Selection 
of evolutionary approach based hybrid data mining algorithms for decision 
support systems an business intelligence”, ICACCI ACM August 2012 
Chennai, India 

[3]  Mrutyunjaya Panda, Ajith Abraham, “Hybrid evolutionary algorithms for 
classication data mining”, Springer, 10 August 2014 

[4]  Rana Forsati, MohammadReza Meybodi, Mehrdad Mahdavi, AzadehGhari 
Neiat. “Hybridization of k means and harmony search methods for web page 
clustering” IEEE International Conference of Web Intelligence and Intelligent 
Agent Technology, 2008 

[5]  Yao Yu, Fu Zhong-liang, Zhao Xiang-hui, Cheng Wen-fang. “Combining 
classier based on decision tree” IEEE International Conference on 
Information Engineering Vol 2. July 2009

[6]  A. Shen, R. Tong, and Y Deng, “Application of classication models on credit 
card fraud detection”, 2007, Service Systems and Service Management, 2007 
International Conference on, pp. 1-4. IEEE

[7]  Kuldeep Randhawa, Chu Kiong Loo, Manjeevan Seera, Chee Peng Lim, 
Asoke K. Nandi, “Credit card fraud detection using AdaBoost and majority 
voting”, 2017, IEEE 

[8]  Suman Arora, Dharminder Kumar, “Selection of Optimal Credit Card Fraud 
Detection Models Using a Coefcient Sum Approach”, International 
Conference on Computing, Communication and Automation (ICCCA2017)

[9]  S Md. S Askari, Md. Anwar Hussain, “Credit Card Fraud Detection Using 
Fuzzy ID3”, International Conference on Computing, Communication and 
Automation (ICCCA2017)

[10]  Luis Vergara, Addisson Salazar, Jordi Belda, Gonzalo Safont, Santiago 
Moral, Sergio Iglesias, “Signal Processing on Graphs for Improving 
Automatic Credit Card Fraud Detection”, 2017, IEEE 

[11]  Fabrizio Carcillo, Yann-Ael Le Borgne, Olivier Caelen and Gianluca 
Bontempi, “An Assessment of Streaming Active Learning Strategies for Real-
Life Credit Card Fraud Detection”, 2017 International Conference on Data 
Science and Advanced Analytics

[12]  Rajeshwari U, Dr B Sathish Babu, “Real-time credit card fraud detection using 
Streaming Analytics”, 2016, IEEE

VOLUME-8, ISSUE-9, SEPTEMBER-2019 • PRINT ISSN No. 2277 - 8160 • DOI : 10.36106/gjra

98 X GJRA - GLOBAL JOURNAL FOR RESEARCH ANALYSIS