Credit Card Fraud Detection Using Unsupervised Machine Learning Algorithms

International Journal of Computer Trends and Technology Volume 69 Issue 8, 1-3, Aug, 2021
ISSN: 2231 – 2803 /doi:10.14445/22312803/IJCTT-V69I8P101 © 2021 Seventh Sense Research Group®
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Credit Card Fraud Detection Using Unsupervised
Machine Learning Algorithms
Hariteja Bodepudi
Irving, USA
Received Date: 01 July 2021
Revised Date: 02 August 2021
Accepted Date: 13 August 2021
Abstract: In the modern era, the usage of the internet has
increased a lot these days and becomes an essential part of
the life. As the e-commerce has increased, the buying and
selling the products over the internet becomes more easy
and flexible. The usage of online shopping, online bill
payment has increased a lot these days with the
introduction of the modern technology like online banking,
credit card payments.
Due to the increase of the online payment and online
shopping, the risk of credit card usage also increased as
the credit card was used in many places as it becomes
hard for the bank to distinguish the real transactions of the
consumer versus the fraud transactions. Also, the credit
card fraud transaction can also happen if the customer
accidentally loses the credit card. So, it becomes complex
for the banks to stop the fraud transactions at that point.
This paper talks about how to detect the fraud transactions
and block the payments before processing by using the
Machine Learning on a real-time basis. This paper clearly
explains about how the anomalies can be detected in an
unsupervised approach and to achieve the higher
accuracy. This Anomaly detection process used in this
paper can be applied in a wide range of applications like
fraud detections in banking, underbilling/overbilling for
customers in telecommunications, security monitoring,
network traffic, health care, and a wide range of
manufacturing industries.
Keywords — Anomaly, Machine Learning, Supervised
Learning, Unsupervised Learning
I. INTRODUCTION
Ecommerce is playing a crucial role in the day-to-day
activities of humans. The Usage of the internet and IOT
devices has increased a lot these days, and that results in
massive volumes of data . [1] As the data increases due to
the collection of data from different forms like structured
and semi-structured data, which is commonly referred to as
Big Data . [2]This large volume of data becomes more
essential for the organizations for the data analysis,
especially for the banks to track the transactions on a real-
time basis. Security of credit cards has become a big
threat in the modern world due to the wide range of online
usage and wide growing data transactions to be analyzed to
detect fraud. It becomes easy for the users to do shopping,
paying bills online in the busy life rather than the physical
visits.
As the usage of credit cards across the globe and online
has increased and data have grown drastically so, it
becomes a challenge for the bank to manually detect the
fraudulent transactions and block the usage. This paper
clearly explains the usage of Artificial Intelligence
Machine Learning Unsupervised Algorithms to detect the
Anomalies.
II. ANOMALY DETECTION
Anomaly detection, which is also referred to as Outlier
Detection, helps us to identify the events, data points that
are far different from the other normal events. These
Anomalies are the data points that deviate from the normal
data points behavior. [3]
Anomalies can be detected in many ways. In this paper,
I’m focusing on detecting the Anomalies, i.e., Fraudulent
Transactions of the Bank, using the Unsupervised Machine
Learning Algorithms.
III. MACHINE LEARNING
It belongs to the branch of Artificial Intelligence . The
concept of Machine Learning algorithms is to learn from
the data, identify the behaviors and predict the future with
the minimal intervention of the mankind. [4]
There are two popular methods for the Machine Learning
widely used across the globe
They are :
1) Supervised Learning Algorithms
2) Unsupervised Learning Algorithms
A. Supervised Learning Algorithms
Supervised Learning Algorithms uses the approach of the
Learning the data patterns using the labeled data, i.e., using
the output. These algorithms are trained with the training
data set, which has both inputs and outputs. The algorithm
is used to learn how the input features are related to the
target variable, i.e., output, in this approach . As the output
is fed into this algorithm, it will adjust the weights until the
model has fitted perfectly . Once the model is trained using
the train data set and that model is used to predict the
dataset, which is unseen before .

Hariteja Bodepudi / IJCTT, 69(8), 1-3, 2021
2
B. Unsupervised Learning Algorithms
Unsupervised Algorithms are the algorithms that are used
for unlabeled data set, i.e., which doesn’t have the output
variable . These algorithms are used to find the unknown
patterns in the data and are used to analyze and segment
them into clusters based on the behaviors. Similar
behavioral patterns of the data are clustered into similar
groups . These algorithms are widely useful for the
unlabeled dataset . [5]
IV. IMPLEMENTATION
In this paper, I’m using the Kaggle credit card data set to
find the fraudulent transactions using the unsupervised
algorithms . [6] So, the credit card data set is not trained
with the output variable, i.e., the target variable . It is
directly trained on the actual dataset without any labels .
After training on the dataset and the next step is to predict
on the same dataset to find the fraudulent transactions in
the whole dataset .
To predict the anomalies in the dataset, I’m used 3
unsupervised algorithms to test the accuracy and see the
best performing algorithm . The reason for using the
unsupervised approach in this paper is mostly in the real
world; there won't be any labeled data available, and also
for the fraud detection, the unsupervised approach is the
best approach .
The algorithms I’m using in this paper are:
1) Isolation Forest
2) Local Outlier Factor
3) One Class SVM
A. Isolation Forest
Isolation Forest belongs to the unsupervised Algorithm .
Isolation Forest algorithm works in the same methodology
as the Random Forest and is constructed on the concept of
decision trees. The main idea and methodology of the
isolation forest algorithm are to detect the anomalies, i.e.,
fraud transactions, instead of using the normal
characteristics of data points. It randomly selects the
features and then randomly splits the feature between min
and max values. [7]
The important parameter in the Isolation Forest Algorithm
is the contamination parameter which we can add the
outlier value like 0.1 if it is 10%, and if we are not sure
about the value can default it to auto for the algorithm to
take care of the outliers while fitting .
Figure1: Isolation Forest [8]
From the figure1, it is clear how the outliers are separated
from the normal data points in the isolation factor . Red
data points that are far from the other data points are
considered as the outliers
B. Local Outlier Factor
Local Outlier Factor (LOF) is an unsupervised algorithm
used in the identification of the anomalies or fraud
transactions in the data . The methodology of the Local
outlier Factor Algorithm measures the density score of
each sample and weighs the scores. It compares the scores
with the other data points, and the lowest score data points
are the anomalies . [9]
Figure 2: Local Outlier Factor (LOF) [10]
From the figure2 we can see the data points and the outlier
scores, and the lower density from the neighbor points are
considered as outliers.
C. One Class SVM
One Class Support vector machine is the unsupervised
algorithm used to detect the outliers in the given dataset .
The idea of the One class SVM is to classify the whole
data into binary classes, i.e., inliers and outliers . It takes
the density of the higher class and the extremity points,
i.e., lower density is considered as outliers . [11]
Figure3: One-Class Support Vector Machine [12]
It is clear from the figure 3 how the One class SVM
separates the outliers . The yellow-colored data points in
the figure are considered as anomalies .

Hariteja Bodepudi / IJCTT, 69(8), 1-3, 2021
3
I trained and predicted the Kaggle credit card data using
the above 3 unsupervised algorithms to identify the
anomalies, i.e., to separate the fraud transactions from the
normal transactions.
V. RESULTS
After predicting with unsupervised algorithms Isolation
Forest, Local Outlier Factor, and One class SVM . The
Isolation Forest outperforms than other 2 algorithms .
A. Accuracy of Model
Algorithm Accuracy
Isolation Forest 99.74%
Local Outlier Factor 99.65%
One-Class SVM 70.09%
Figure 4: Results Table from the Algorithms
Figure5: Results Plot from the Algorithms
From the above figure4 and figure5 Results table and plot ,
we can see the accuracy of Isolation Forest is 99.74%,
Local Outlier Factor is 99.65%, and One-Class SVM is
70.09%.
So, when compared with all the 3 models, results of
Isolation Forest and Local Outlier Factor Algorithms are
almost performing similarly and better than the One class
SVM, which is poorly performing .
VI. CONCLUSION
This paper talks about how the credit card fraudulent
transactions can be detected with the unsupervised
algorithms. In the real world, we won’t be having the
outputs for the inputs, so this paper concentrates on
unsupervised learning so that way these algorithms can
detect the anomalies without even observing the output
variables .
Unsupervised algorithms are very useful and good
detectors of the anomalies . The algorithms used in this
paper can be applied to any field for anomaly detection,
i.e., to detect the outliers or fraudulent transactions .
VII. REFERENCES
[1] H. Bodepudi., Faster The Slow Running RDBMS Queries With
Spark Framework, (2020). [Online]. Available:
https://www.researchgate.net/publication/347390599_Faster_The_
Slow_Running_RDBMS_Queries_With_Spark_Framework.
[Accessed 07 2021].
[2] H. Bodepudi., Data Transfer Between RDBMS and HDFS By
Using The Spark Framework In Sqoop For Better Performance,
Internation Journal of Computer Trends and Technology, 69(3)
(2021) 1.
[3] Anaomaly-Detection., [Online]. Available:
https://avinetworks.com/glossary/anomaly-detection/. [Accessed
July 2021]., (2021).
[4] What is Machine Learning., [Online]. Available:
https://www.ibm.com/cloud/learn/machine-learning. [Accessed
July 2021].
[5] Popular Machine Learning Methods, [Online]. Available:
https://www.sas.com/en_us/insights/analytics/machine-
learning.html. [Accessed July 2021].
[6] CreditCard Data, [Online]. Available:
https://www.kaggle.com/mlg-
ulb/creditcardfraud?select=creditcard.csv. [Accessed July 2021].
[7] H2o.ai Data Science, ,[Online]. Available:
https://docs.h2o.ai/h2o/latest-stable/h2o-docs/data-science/if.html.
[8] Isolation Forest Example, [Online]. Available: https://scikit-
learn.org/stable/auto_examples/ensemble/plot_isolation_forest.htm
l#sphx-glr-auto-examples-ensemble-plot-isolation-forest-py.
[9] Anaomaly Detection With Local Outlier Factor, [Online].
Available: https://www.datatechnotes.com/2020/04/anomaly-
detection-with-local-outlier-factor-in-python.html.
[10] Outlier detection with Local Outlier Factor (LOF), [Online].
Available: https://scikit-
learn.org/stable/auto_examples/neighbors/plot_lof_outlier_detecti
on.html. [Accessed August 2021].
[11] One-Class Support Vector Machine., [Online]. Available:
https://www.xlstat.com/en/solutions/features/1-class-support-
vector-machine.
[12] One-Class SVM.M [Online]. Available: https://scikit-
learn.org/stable/auto_examples/svm/plot_oneclass.html#sphx-glr-
auto-examples-svm-plot-oneclass-py.
99.74% 99.65%
70.09%
ACCURACY
Accuracy of the Model
Isolation Forest Local Outlier Factor
One-Class SVM

Credit Card Fraud Detection Using Unsupervised Machine Learning Algorithms

More Related Content

What's hot (20)

Similar to Credit Card Fraud Detection Using Unsupervised Machine Learning Algorithms (20)

Recently uploaded (20)

Credit Card Fraud Detection Using Unsupervised Machine Learning Algorithms