Volume 9, Issue 3, March – 2024 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24MAR961

Credit Card Fraud Detection

Shakuntala D. S.1; Shreyansh Kuchanur2; Smitha M. P.3; Sumanth J. M.4; Dr. Kavitha C.5
1,2,3,4
Student; 5Head of Department,
Department of Computer Science and Engineering,
Dayananda Sagar Academy of Technology and Management, Bangalore, Karnataka, India

Abstract:- While credit card fraud and abuse are the way I. INTRODUCTION
becoming more common, the convenience using of there
credit for the way online purchases has also improved. A. An Overview of Credit Card Fraud Detection
These fraudulent activities pose a severe financial danger to Unauthorized transactions done by someone other than the
both credit the are card users and way they using it financial cardholder using credit card information that has lost, stolen, or
institutions. The first thinking aims to do of this research falsified are known as credit card fraud. Card-not-present fraud,
project is to recognize and put an end to these kinds of in which credit card credentials are stolen and used to make
fraudulent activities. It addresses a broad range of subjects, online purchases, has increased in frequency due to the growth
including the frequency of false positives, imbalanced of internet shopping. Every year, fraudulent actions rise as of
datasets, evolving fraud trends, and restricted public data result of the thought widespread use these and of systems and
access. e- banking, causing large financial losses.

The literature now under publication offers a range of The risk of credit way is card fraud has increased due to
machine learning-based methods, including logistic the widespread usage of credit way cards for l transactions, in
regression, decision trees, random the forests, support to the age of digital commerce. Due to the possibility of identity theft
vector of the +machines, and XG Boost, with the purpose of and large financial losses, credit way card fraud is the a serious
identifying credit card fraud. However, these methods often concern to financial individual cardholders.
exhibit lower accuracy rates, highlighting the need for more
advanced deep learning algorithms in order to effectively The incorporation of machine way of learning algorithms
lower fraud losses. Therefore, the of way they primary for credit way card fraud detection has become a crucial tactic
thinks objective of this is to improve fraud detection in response to issues.More complex way of conventional rule-
abilities through which the way this application most be of based methods in the field of to the credit card fraud detection.
state- of-the-art deep learning algorithms. Methods including support vector machines (SVMs), random
forests, logistic regression trees, and neural networks have been
An assessment way it should be of the research project extensively employed to develop predictive models that can
reveals better outcomes, including optimized AUC curves, differentiate between authentic and fraudulent By training
precision, f1- score, and accuracy. models on past transaction data, machine of learning detect
credit card fraud by teaching algorithms to recognize patterns
The ultimate objective there are is to create models that point to fraudulent activity. These models are more
that will greatly improve credit card fraud detection and successful at identifying previously undetected fraudulent
prevention. This research focuses on advanced deep conduct because they are always changing and reacting to new
learning techniques in to the way it is order they provide the fraud strategies. By reducing the amount of false positives, the
to more reliable and accurate fraud detection mechanisms, integration of machine learning technologies helps minimize
enhance security for credit card users, and lower financial the disruption of valid transactions while simultaneously
risks for financial institutions when conducting online increasing the accuracy of fraud detection. To preserve their
transactions. efficacy, machine learning models must be continuously
developed and improved due to the ongoing growth of
Keywords:- Transaction Data Analytics, Online Fraud, Credit fraudulent techniques.
Card Fraud, Deep Learning, Machine Learning, Fraud
Detection. The in case use of several machine way they learning
techniques and of the algorithms for credit fraud detection is
examined way there are in this article. It covers the
fundamentals of these methods, their benefits, drawbacks, and
possible applications in the they are been area of credit the card
for machine learning- based prevention of this enhanced
security.

IJISRT24MAR961 www.ijisrt.com 854

 Volume 9, Issue 3, March – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24MAR961

II. BRIEF OVERVIEW OF CREDIT CARD FRAUD G. Data Distribution

DETECTION USING MACHINE LEARNING One typical use of imbalanced binary classification is the
TECHNIQUES identification of fraudulent way the credit card transactions,
where the emphasis positive class (fraud) and negative class
A. Machine Learning-Based Way for Credit Card Fraud (non-fraud).By displaying the positive and negative
Detection distributions, the issue of class imbalance is better understood.
Numerous number of the machine way of learning
methods, such as the Extreme Learning way the Method, H. Machine Learning Algorithms
Decision, Random the Forest, Support Vector Machine, A Comparative Study Using accuracy and F1 measure
Logistic Regression, and XG Boost, have that been used to metrics, a comparative analysis of applied ML algor for credit
identify credit card fraud. The limited accuracy of current card fraud
machine learning methodologies necessitates the development
of cutting-edge deep way of things learning algo to minimize III. REVIEW OF PAPER 1
fraud losses.
A thorough investigation way the use of sophisticated
B. A Comparative Way of Analysis of Machine the Learning machine of the learning (ML) and deep learning (DL) credit card
and Deep Learning Algorithms of fraud (CCF) detection is presented in the paper "Credit Card
Convolutional a of neural network (CNN) architectures Fraud detection is conducted to shed light how well various
are implemented after machine learning methods have been models work with an emphasis on fraudulent purchases. a
applied empirically to detect fraud using a reference data set of summary of the assessment metrics, the dataset, and the results
European cards. This improves the performance of fraud of tests conducted on an actual dataset. The study is broken up
detection. Evaluation of the study demonstrating enhanced into several sections, including an analysis of pertinent
AUC outcomes. literature, a detailed description of the proposed model and how
it works, The paper also discusses the distribution of way the
C. Description of the Way the Credit Card Data Set this credit card transaction of the data, the architecture of the
During the course of two days, cardholder transactions suggested CNN model, and the range of training epochs[1].
totaling 284,807 were included in the credit card data set used Furthermore, the study highlights the effectiveness of deep
for research. Of these, 492, or 0.172 percent, were fraudulent. learning algorithms for CCF categorization, including CNN and
Most of the dataset's features are subjected to principal of the LSTM, and offers insights Exploring the significance of pre-
component the way analysis (PCA), which reduces processing data for machine learning[1]. Overall, the study
dimensionality, improves interpretability, and minimizes provides a thorough which analysis of the use of sophisticated
information loss. ML and DL algorithms for the detection means of credit way
they card fraud, offering insightful information about the
D. Top 10 Machine Way they Learning Algorithms for Fraud efficacy and performance they of the suggested CNN model
Detection relative to other methods.
To detect credit card fraud, various algorithms are used,
including Linear Regression, Logistic Regression, Decision A. Merits:
Tree, SVM, Naïve Bayes, CNN, K-Means, Random Forest,
Dimension Reduction Algorithms, and Gradient Boosting  Comprehensive Analysis: Using machine way of learning
Algorithms. (ML) and deep way of learning (DL) algorithms, the paper
offers a thorough credit card the way fraud detection. It also
E. Model Confusing Metrics includes a thorough explanation of the various evaluation
The confusion metric, a representation of a classification metrics and approaches.
model, shows how well the model is projected to the outcomes  Application of State-of-the-Art methods: The study
that were previously linked to the early ones. The confusion shows a dedication to utilizing cutting-edge technology for
metrics can be visualized by using the association table as a useful applications by applying state-of-the-art deep
heatmap. learning way of methods, such a way that as convolutional
they are neural networks (CNN), to increase fraud detection
F. Machine Learning Algorithms' Accuracy performance.
There are six different types of structured classification
 Comparative Analysis: In addition to comparing the
models, including well-known models are the way like logistic
performance of the suggested CNN model to other methods,
regression, decision trees, and support vector machines. The
the publication also offers a comparative analysis of both
accuracy which means the exact of the classifiers is shown by
ML and DL algorithms, offering insightful information
the reported things are results of applied machine learning
about the subject of credit card the fraud detection.
techniques.

IJISRT24MAR961 www.ijisrt.com 855

Volume 9, Issue 3, March – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24MAR961

B. Demerits: fraud detection problems.

 Lack of Data Disclosure: The topic of concealment is A. Merits:

brought up in the document, which may limit the
transparency and repeatability of the research findings by  Superior Performance: Comparing the suggested LSTM
preventing the publication of the structures and background. ensemble with SMOTE-ENN to other cutting-edge
Detection Using State-of-the-Art Machine Learning and techniques, the ensemble performed exceptionally well,
Deep Learning Algorithms"[1]. The study compares a demonstrating the method's resilience.
convolutional of the neural network (CNN) model's  Effective Data Resampling: The of the SMOTE-ENN data
performance analysis to that of other ML and DL resampling technique was demonstrated by the study, which
techniques already in use. The study places a strong also revealed that the resampled data considerably improved
emphasis on evaluating the efficacy of the classifiers using the performance of the different classifiers, including the
performance evaluation metrics like accuracy, precision, suggested ensemble.
and recall[1]. The experimental transactions made by  Robust Solution: An important advancement in the way
cardholders across Europe, these field is the creation of a reliable technique way these
 Ethical and Privacy Considerations: The use of credit card are been for detecting credit card fraud utilizing an LSTM
transaction data the for research purposes may have ethical neural network ensemble, in addition to efficient feature
ramifications and raise privacy concerns, which the study engineering through the resampling of unbalanced data.
does not specifically address. This is a crucial point to take  Adaptation to Dynamic Shopping Behavior: By adjusting
into account when discussing data security and privacy. to the changing shopping habits of credit card customers, the
 Real-World Implementation Challenges: The document AdaBoost methodology, which creates strong classifiers
may not address the potential challenges associated with the less prone to overfit, and the LSTM algorithm, a reliable
real-world. method for modeling sequential data, combine to provide an
efficient way to detect credit way of finding card fraud.
IV. REVIEW OF PAPER 2  High Predictive Ability: With a ROC curve that is to the
upper-left corner and a higher AUC value than other
A neural way the learning network the way ensemble of classifiers, the suggested LSTM ensemble performed well
classifier method its are and a hybrid data technique are used in in identifying both way leaning fraudulent and valid
the collection study "A Neural Network Ensemble With Feature transactions.
Engineering for Improved Credit way of the Card the Fraud
Detection" to provide an effective method of credit that are card B. Demerits:
fraud detection[2]. The paper tackles the problems of
unbalanced credit card datasets and the static mapping of input  Poor Sensitivity without Data Resampling: When trained
vectors to output vectors that renders traditional machine of with the original, un resampled data, the classifiers
they learning methods ineffective for detecting fraud. The including the suggested ensemble obtained low sensitivity
suggested way of the approach makes usecase way they of the scores, demonstrating the effective way they of class
edited nearest neighbor (SMOTE-ENN) method for data imbalance on performance.
resampling and the synthetic minority oversampling  Class Imbalance Challenge: Conventional machine of the
methodology and long s network as the base learner in the way learning methods are challenged by the inherent class
adaptive boosting (AdaBoost) technique[2]. The experimental imbalance in credit way finding card the transaction
findings show that LSTM ensemble performs better than datasets, which results in poor sensitivity and
alternative benchmark techniques, reaching a high way misclassifications.
understanding level of specificity and sensitivity[2]. The way it  Complexity of Data Resampling: Despite its
is suggested method's higher effectiveness is demonstrated by effectiveness, the SMOTE-ENN data resampling technique
a comparison with current way of credit way of card fraud may complicate the fraud way they are detection system's
detection learning techniques in the paper[2]. Furthermore, the preprocessing step, necessitating cautious adoption and
study shows the robustness of the suggested method in oversight.
identifying means fraudulent way of things which are  Limited Comparison Metrics: The study may have
transactions and proves its efficacy using real-world credit way ignored other crucial measurements like precision,
of card transaction datasets[2]. The that was employed, the accuracy, and F1 score by concentrating mostly on
adaptive boosting algorithm, and the classifiers' performance sensitivity, specificity, and AUC as performance evaluation
both with and without data resampling are all covered in the metrics.
study. It offers a thorough which they are been analysis of the  Dataset Limitations: The research employed publicly
current approaches to dealing with unbalanced data in credit accessible real-world credit the way card transaction
card fraud the way detection and emphasizes the of deep way datasets, which might not accurately capture the variety and
learning methods, especially recurrent neural networks, to solve intricacy of credit card transactions in various locations or

IJISRT24MAR961 www.ijisrt.com 856

Volume 9, Issue 3, March – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24MAR961

situations, there by restricting the applicability of the  Incorporation of Diversity Measure: The approach takes
suggested methodology. into account the variety among weak classifiers, calculating
diversity by introducing a double-fault measure and adding
V. REVIEW OF PAPER 3 it to the weak learners' weight computation. This improves
the model's ability to generalize and increases its efficacy in
The usage of adaptive enhancement (Adaboost) models for detecting fraud.
fraud detection and the rising incidence of credit card fraud  Practical Applicability: It is demonstrated algorithm of the
occurrences in online transactions are covered in the document. way performs better than alternative Adaboost
It the way how we solves the classic Adaboost algorithm's lack improvement techniques in of the terms of F1 value and
of resilience, which is mostly caused by the basic classifier AUC value, indicating its usefulness in practical settings.
selection strategy's exclusive focus on error rate[3]. In order to This implies that the technique can be applied successfully
improve the Adaboost algorithm's objective function, the study in the detection of credit of the card fraud and possibly in
suggests D- AMWSPLAdaboost, an enhanced Adaboost other fields with related problems.
algorithm that combines an adaptive hybrid weighted self-
paced learning technique. It also integrates a double-fault B. Demerits:
metric into the computation of weak learners' weight and
introduces it as a way to measure variation among base  Underfitting Concerns: Accordingway they are to the
categories[3]. AUC value and F1 theway which think suggested research, underfitting may occur in the suggested AMWSPL
through the algorithm performs better than previous Adaboost Adaboost algorithm when the number maximum way of
enhancement algorithms when tested on credit card fraud iterations is low. This implies that selecting the number
datasets[3].The significance of performance evaluation which it means of iterations for practical application
measures in unbalanced classification problems—specifically, requires significant thought.
credit card fraud detection—is also covered in the research,  Complexity: The algorithm becomes more sophisticated
including AUC and F1. According to experimental results, D-
with the team addition of diversity metrics and adaptive
AMWSPL Adaboost performs better overall and converges
hybrid weighted self-paced learning. Although these
mostly the quickly than other algorithms in terms of AUC and
improvements lead to better performance, there's a chance
F1 values[3]. The suggested way of the algorithm offers a more that they also make understanding and implementation more
reliable and efficient method way they for identifying credit
difficult.
card fraud, and it has the potential to be put to good use in actual
 Limited Comparison: The suggested approach is
situations.
compared in the paper to previous Adaboost improvement
algorithms, but more comparisons with a wider variety of
A. Merits:
fraud detection techniques would be beneficial to give a
more thorough evaluation of its efficacy.
 Enhanced Robustness: The adaptive hybrid weighted
 Lack of Real-world Implementation Results: Although
selfpaced learning technique included in the suggested
the approach performs well in experimental conditions, the
AMWSPLAdaboost algorithm solves the weak robustness
publication lacks case studies or detailed real-world
of the conventional Adaboost algorithm. This improves the
implementation findings that could further support the
classifier's overall performance and convergence speed,
algorithm's applicability.
increasing its efficacy in detecting credit way they card
 Algorithm-Specific Limitations: Certain restrictions or
fraud.
constraints, like as computational efficiency, scalability to
 Improved Performance Metrics: As assessment metrics,
bigger datasets, or adaptability to emerging fraud trends,
the method makes use of AUC and F1 values, which are
may apply to the things proposed approach and are not
especially well-suited for unbalanced classification issues
completely addressed in the work. These factors ought to be
such as credit way thinking card fraud detection. This makes
taken into account for a more thorough assessment of the
it possible to evaluate the model's performance in-depth,
algorithm's suitability.
producing outcomes that are more accurate and trustworthy.
 Effective Handling of Imbalanced Data: In order to
VI. REVIEW OF PAPER 4
produce more accurate and superior findings, the research
focuses on processing imbalanced data by utilizing The paper "Performance Evaluation of way Machine
undersampling strategies and other machine learning Learning Methods for Credit Card Fraud Detection Using
methods. This method is essential for detecting from credit
SMOTE and AdaBoost" provides a comprehensive analysis of
way the card fraud since it helps identify fraudulent
the use case of the of ML algorithms for credit way which card
transactions, which can be difficult to distinguish from fraud detection[4]. The study addresses the issue of class
authentic ones. imbalance in real-world datasets and evaluates the effectiveness
of various machine learning techniques, including the
combination of Decision Tree, Extra Tree, Support Vector

IJISRT24MAR961 www.ijisrt.com 857

Volume 9, Issue 3, March – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24MAR961

Machine, Random Forest, Extreme Gradient Boosting, and B. Demerits:

Adaptive Boosting (AdaBoost)[4]. The work focuses on using
the method may be Syntheticis is the Minority Over-sampling  Limited Real-World Dataset Validation: Although the
Technique (SMOTE) to address class imbalance and combines suggested framework is validated on a synthetic dataset by
machine way the learning (ML) techniques with AdaBoost to the research, it has not been validated on a large-scale real-
increase performance[4]. The framework is validated using a world credit card fraud dataset, which could limit the
highly skewed synthetic credit way of card fraud dataset, results findings' applicability in real-world scenarios.
demonstrate that AdaBoost improves such a way learning the  Focus on Specific ML Methods: The study's primary focus
performance of the ML approaches. In way terms of accuracy, is on assessing particular machine learning techniques,
recall, precision, Matthews Correlation Coefficient (MCC), and which may restrict the investigation of other algorithms that
Area of the Under the Curve (AUC), applying AdaBoost could enhance the efficacy of credit way the card fraud
enhances the trial findings. Additionally, the study detection.
demonstrates that the which are been proposed framework  Lack of Exploration of Additional Performance
outperforms existing ML-based credit way of the card fraud Metrics: The study mainly concentrates on the performance
detection algorithms in terms of fraud detection accuracy[4]. In- parameters of accuracy, recall, precision, Matthews
depth examination of past research using machine way the Correlation Coefficient (MCC),possibly ignoring other
learning techniques to identify credit The report also covers pertinent indications that could offer a more thorough
card fraud, which provides insight into the field's current assessment of fraud detection models.
state.Taking everything into consideration, the research  Limited Discussion on Model Interpretability: The
contributes to the creation of a scalable framework that interpretability of the way the such machine way the are
effectively corrects class imbalance and enhances classification learning models used for fraud detection is not sufficiently
quality by utilizing AdaBoost in credit way it is card fraud covered in the study, despite the fact that it is crucial to
identification[4]. The results show that there is room for comprehending the decision-making process and
improvement in accuracy and validate the value of thesuch way guaranteeing openness in credit way card the way fraud
proposed framework. detection systems.
 Overemphasis on Synthetic Dataset Validation:
A. Merits: Although the synthetic dataset validation is useful, a more
thorough investigation of real-world data validation would
 Scalable Fraud Detection Framework: In to they are they help the research to assure the robustness and practical
order to detect fraudulent activity in credit of they card application of the suggested methodology.
transactions, the research offers a scalable methodology for
detecting credit way card fraud that effectively solves class VII. REVIEW OF PAPER 5
such that imbalance in real-world datasets.
 Implementation of SMOTE Technique: In way they are The difficulty of imbalanced datasets in credit of the card
been order to solve such way of the the problem of class are fraud detection is addressed in a novel way in the research
totally imbalance in credit way of card fraud datasets, the "Credit Card Fraud Detection Based on Improved Variational
study uses the Synthetic Minority Over-sampling Technique Autoencoder Generative Adversarial Network"[5]. The work
(SMOTE), which shows how to handle skewed data suggests an enhanced VAEGAN model for data augmentation
distributions in an aggressive manner. and evaluates the effectiveness of several oversampling
 Ada Boost Method Integration: The study combines techniques, such as SMOTE, GAN, VAE, and VAEGAN. The
various machine learning techniques with the Adaptive experimental results show that the modified VAEGAN
Boosting (Ada Boost) algorithm to improve classification technique is useful for improving the classification model's
quality and performance, demonstrating a thorough strategy precision, recall, and F1_score—especially when handling such
for boosting fraud are the were detection model efficacy. an are imbalanced data classification issues.'Time' and 'Amount'
 Comparative Analysis of ML Methods: Using a publicly features, together with numerical features that have undergone
accessible dataset of actual credit card transactions, the PCA dimensionality reduction, are among the 30 features in the
study compares and contrasts a number they are been used credit way that card fraud dataset that the study uses from the
of machine learning techniques, offering insightful Kaggle platform[5]. The dataset presents a typical imbalanced
information about the efficiency and performance of various classification problem because it is severely imbalanced, with
fraud detection systems. only 492 fraudulent transactions out of 284,807 total
 Validation on Synthetic Dataset: The suggested structure transactions. To solve this problem, the authors rebalance the
is verified using an extremely unbalanced artificial credit training set by using oversampling techniques and normalizing
way of the card fraud dataset, verifying the efficacy of the the transaction amount[5]. Based on its superior performance,
study findings and showcasing its potential for practical the XGBoost classification algorithm is chosen as the baseline
implementation. model for fraud detection, according to the experimental
results. The study shows how well the suggested oversampling

IJISRT24MAR961 www.ijisrt.com 858

Volume 9, Issue 3, March – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24MAR961

strategy works to improve the classification model's F1_score  Limited Generalizability: Although the study shows that
and precision by contrasting the performance of the XGBoost the suggested strategy works, more testing and validation
algorithm with that of other machine way of the learning and may be necessary before applying it to other fraud detection
deep way of learning techniques[5]. The study concludes with datasets and real-world applications.
a thorough analysis of oversampling techniques and how they  Impact of Data Imbalance: The study draws attention to
affect the identification of credit way card fraud. The suggested the issue of imbalanced way data sets in credit which card
enhanced VAEGAN approach provides a dependable and fraud detection and makes the case for the necessity of
efficient solution for real-world fraud detection applications by ongoing investigation and the creation of reliable remedies.
demonstrating encouraging outcomes in strengthening the  Potential Trade-Offs: The findings of the experiments
resilience and efficacy of the classification model[5]. The paper suggest possible trade-offs between various resampling
makes a significant contribution to the fields of information techniques and classification performance indicators,
security and machine of which learning by offering insightful necessitating careful evaluation of particular needs and
solutions for dealing with unbalanced datasets in fraud way objectives in fraud most detection applications.
detection.
VIII. CONCLUSION
A. Merits:
To sum up, creating and putting into place reliable
 Enhanced Precision and F1_Score: In particular, the mechanisms for detecting credit way of the card fraud is
enhanced VAEGAN oversampling approach outperforms essential to safeguarding financial transactions. Sophisticated
the original one in improving the classification model's technologies like machine learning, data analysis, and pattern
accuracy and F1_score while handling unbalanced data recognition significantly enhance the speed at which fraudulent
classification issues. activity can be detected. To keep abreast of developing fraud
 Effective Classification Results: The efficacy of the strategies, detection systems must be continuously improved
suggested strategy is demonstrated by the experimental and adjusted. Financial institutions, academics, and regulators
findings, which show that the XGBoost classifier combined must work together to promote innovation and successfully
with the enhanced VAEGAN algorithm produces superior counter new dangers. The adoption of a proactive strategy that
classification results than other classifiers. incorporates advanced detection methods is necessary in which
 Comprehensive Assessment: The paper offers a thorough they order to reduce risk, preserve trust in financial institutions,
analysis of resampling techniques and how they affect credit and guarantee secure transactions for all stakeholders.
card fraud detection, providing insightful information for
dealing with unbalanced data sets in fraud detection. REFERENCES
 Efficiency and Robustness of the Model: The suggested
resampling technique offers a dependable and effective [1]. Raj S.B.E., Portia A.A., An Analysis of Credit Card
solution for real-world fraud detection applications while Fraud Detection Methods, International Conference on
also significantly enhancing the classification model's Computer, Communication and Electrical Technology
robustness and efficiency. (ICCCET) (2011), 152-156.
 Benchmarking Analysis: The work presents a thorough [2]. Jain R., Gour B., Dubey S., A hybrid approach to credit
grasp of the efficacy of resampling techniques by comparing card fraud detection using rough set and decision tree
the performance of several approaches, such as GAN, VAE, techniques, International Journal of Computer
and SMOTE, and shows that the enhanced VAEGAN Applications 139(10) (2016).
method performs better in terms of accuracy, F1_score, and [3]. Dermala N., Agrawal A.N., Credit card fraud detection
other metrics. the method. using SVM and false alarm reduction, International
Journal of Innovations in Engineering and Technology
B. Demerits (IJIET) 7(2) (2016).
[4]. Hafiz K.T, Aghili S, Zavarsky P, Using predictive
 Shortcomings: Recall reduction possibility: In certain analytics technology to detect credit card fraud, 11th
instances, the enhanced VAEGAN method's recall is Iberian Conference on Information Systems and
marginally lower than that of alternative approaches, Technologies (CISTI) (2016), 1.
suggesting a possible accuracy vs. recall trade-off that [5]. Sonepat H.C.E Survey Paper on Credit Card Fraud
should be carefully evaluated. Detection, International Journal of Advanced Research
 Model Training Complexity: Resampling techniques and in Computer Engineering & Technology (2014).
ensemble learning classification algorithms can complicate [6]. R. Wheeler, S. Aitken, "Multiple algorithms for fraud
model installation and training, necessitating careful detection" in Knowledge-Based Systems, Elsevier, vol.
consideration of training time and computer resources. 13, no. 2, pp. 93-99, 2018.

IJISRT24MAR961 www.ijisrt.com 859

Volume 9, Issue 3, March – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24MAR961

[7]. S. Patil, H. Somavanshi, J. Gaikwad, A. Deshmane, R.

Badgujar, "Credit Card Fraud Detection Using Decision
Tree Induction Algorithm", International Journal of
Computer Science and Mobile Computing (IJCSMC),
vol. 4, no. 4, pp. 92- 95, 2020, ISSN ISSN: 2320-088X.

IJISRT24MAR961 www.ijisrt.com 860