2024 2nd International Conference on Disruptive Technologies (ICDT)

Secure UPI: Machine Learning-Driven Fraud

Detection System for UPI Transactions
Rupa Rani Adnan Alam Abdul Javed
Department of Computer Science & Department of Computer Science & Department of Computer Science &
Engineering, Ajay Kumar Garg Engineering- Data Science, ABES Institute Engineering- Data Science, ABES Institute
Engineering College, Ghaziabad, of Technology, Ghaziabad, U.P., India. of Technology, Ghaziabad, U.P., India.
Ghaziabad, U.P., India. Email: [email protected] Email: [email protected]
Department of Computer Science &
Engineering, Banasthali Vidyapith,
Rajasthan
Email: [email protected]
ORCID ID- 0000-0002-4783-3857

Abstract— SecureUPI specializes in developing an advanced methods that comply with felony requirements and industry
2024 2nd International Conference on Disruptive Technologies (ICDT) | 979-8-3503-7105-5/24/$31.00 ©2024 IEEE | DOI: 10.1109/ICDT61202.2024.10489682

fraud detection gadget the usage of the effective XGBoost device first-rate practices are used to shield and anonymize the
getting to know set of rules to create an advanced fraud identity statistics that has been gathered. The system is made to
device. XGBoost is a properly-proper alternative for enhancing paintings in the constraints of privateness and statistics safety
the precision of fraud detection fashions because of its laws, selling person self assurance and criminal compliance
reputation for dealing with tricky datasets with efficiency and [5].
its music record of success throughout multiple industries. In
order to extract pertinent records, like transaction quantity,
frequency, and place, our approach preprocesses UPI
transaction facts. This article makes use of a labelled dataset to
train the XGBoost version in order that we may also take gain
of its sturdy prediction talents and capacity to handle
imbalanced datasets. To help create a system that is less difficult
to apprehend and use, feature importance evaluation is used to
discover essential symptoms of feasible fraud. After training, the
model is covered right into a real-time UPI transaction tracking
device, in which it maintains an eye fixed out for any suspicious
traits in incoming transactions. In order to lessen the results of
fraudulent activity, the system is constructed with 98.2 %
accuracy to send out instant notifications and take preventive
steps. This challenge allows in improving UPI transaction   
security and advancing economic era are accomplished through
demonstrating the performance of machine learning in fraud Fig. 1. Interface of Secure UPI
detection.
Figure 1 shows the interface of Secure UPI website which
Keywords— UPI, Fraud Detection, SMOTE, PCA, XGBoost, is supposed to be advanced constantly it isn't static [6] This
Imbalanced Datasets, Dimensionality Reduction, Machine structure of the device consists of non-stop mastering, which
Learning. lets in it to alter to converting fraud techniques and customer
conduct. By mechanically retraining the version, the device
I. INTRODUCTION makes use of the most modern records to beautify its
In a global in which virtual transactions rule, the Unified predicting capabilities. By incorporating non-save you
Payments Interface (UPI) has become a current platform that studying, the UPI Fraud Detection System is assured to remain
streamlines and expedites monetary transactions [1]. But the a sincere protector within the path of latest threats and to hold
quick development of digital bills also gives the undertaking its efficacy over the years [7].
of fighting state-of-the-art fraudulent sports activities.
Through the advent of a sophisticated UPI Fraud Detection For the cause of resolving the inherent elegance imbalance
System and the utility of current pre-processing, feature in UPI transaction datasets, the Synthetic Minority Over-
extraction, and algorithm choice techniques, this venture sampling Technique (SMOTE) choice made for the duration
tackles this essential assignment [2] [3]. of pre-processing is essential [8] [9]. Since fraud times are
commonly unusual, SMOTE creates fictitious examples to
Improving virtual transaction security and dependability even out the commands just so the following set of rules can
via the UPI framework is the principle intention of the UPI teach on a greater representative dataset. By doing this, the
Fraud Detection System. The tool pursuits to hit upon and model's capacity to identify fraudulent patterns is improved
decrease fraudulent sports activities via using modern and bias against the majority class is eliminated. [10] Because
techniques, making certain the safety and reliability of the UPI Principal Component Analysis reduces dimensionality while
surroundings [4]. During the UPI Fraud Detection System's effectively collecting the most important information in the
design and deployment, protection and privateness of dataset, it is used. principle component analysis (PCA) helps
sensitive economic records are of utmost importance. Strict identify patterns that significantly contribute to the variation

979-8-3503-7105-5/24/$31.00 ©2024 IEEE 924

Authorized licensed use limited to: ANURAG GROUP OF INSTITUTIONS. Downloaded on August 01,2024 at 09:05:56 UTC from IEEE Xplore. Restrictions apply.
 2024 2nd International Conference on Disruptive Technologies (ICDT)

in the data by converting the original features into set of policies, which serves because the primary class model.
uncorrelated principle components. As a result, features are XGBoost comes with severa hyperparameters, together with
represented more effectively, which is crucial for developing studying fee, most depth of wooden, extensive form of
reliable fraud detection models [11]. boosting rounds, and regularization phrases, among others
[18].
The UPI Fraud Detection System operates in real-time,
continuously monitoring incoming transactions as they occur The system has thorough reporting and assessment
[12]. By utilizing the XGBoost algorithm's speed and capabilities that offer facts about effectiveness, overall
efficiency, transactions are quickly compared to the learnt performance, and abnormalities determined [19].
patterns, enabling the immediate detection of potentially Stakeholders can advantage a complete records of the fraud
fraudulent activity. By ensuring that any suspicious activity is detection panorama, discover the facts of flagged transactions,
immediately addressed, real-time monitoring helps to reduce and realise the features that make contributions to danger
the negative effects of fraudulent transactions on both users scores via the use of interactive dashboards and visualizations.
and financial institutions [13]. This reporting feature permits with compliance reporting,
enables well-knowledgeable preference-making, and gives
insightful enter for brought machine development [20]
In cease, a high development round economic safety is the
UPI Fraud Detection System. Using behavioural evaluation,
anomaly detection, and tool studying, the challenge seeks to
gather a strong defence in opposition to fraudulent UPI
transactions. This all-encompassing method not exceptional
identifies and stops fraudulent interest, however it additionally
Fig. 2. Data Splitting
advances protection capabilities constantly inside the ever-
Figure 2 shows how we are splitting the data into 2 changing region of virtual banking. The undertaking has the
sections. Then we will search for the best parameter functionality to noticeably improve economic safety by means
combination for the XGBoost algorithm, using K-fold cross of the use of strengthening client don't forget and the of upi
confirmation. Before we get into the details of parameter transaction.
optimization, a caveat about how I use data for cross II. LITERATURE SURVEY
confirmation is in order. XGBoost offers a useful point called
beforehand stopping, which helps help overfitting due to M.A Ibrahim [1], This research paper discusses the most
growing too numerous trees. At every replication, it evaluates common methods of fraud, detection techniques, and recent
the performance of the model so far on a confirmation set. It findings in the field. The researchers used the SMOTE
keeps training until the error on this confirmation set, which technique to balance the dataset and found that models like
I'll call beforehand stopping set, fails to ameliorate for a given Decision Tree, Random Forest, Neural Network, and K-
number of duplications. This means that we're optimizing the nearest neighbour performed well when fitted and trained with
number of trees as we train the model. But that also means that the data. The system allows users to select their preferred
we can not use that same set to estimate the model. else, we'd model. The Random Forest model achieved an accuracy of
overfit( the number of trees parameter) and the cross 93.58%, but its efficiency decreases when trained with
confirmation score will be poisoned overhead. That's why we imbalanced transaction datasets.
need a separate confirmation set for early stopping, in addition P. Boulieris [2], The main objective of the work is to
to the( rotating) confirmation set used in cross confirmation. present a dataset for online fraud detection that is anonymized
thus, to help the early stopping point from overfitting, in and publicly available. We argue that standard evaluation
addition to unyoking the data into a training and a test set, I metrics used in existing literature should be complemented
also set aside a small portion of the training set to be used as with online and offline assessments to evaluate model
the early stopping confirmation set. performance in a real-world business setting. We found that
The primary classification model is XGBoost, an ensemble incorporating anomaly detection features improves all metrics
learning algorithm selected for its outstanding performance except for online detection, highlighting the importance of
and adaptability. It is especially well-suited for fraud detection considering both online and offline evaluation alongside
due to its capacity to manage intricate linkages, nonlinearities, standard metrics. Despite using fewer traditional features, the
and interactions within the data base. An extremely precise addition of NLP-based features significantly improved
and flexible classification procedure is guaranteed by the performance compared to a previous study.
ensemble of decision trees, which have been trained B. Baesens [9], This paper discusses the importance of
successively to fix the mistakes of the earlier models [14] [15]. information technology in fraud detection. The authors share
The project consists of an in depth hyperparameter information technology for feature design and application
tweaking device to get the most predictive power out of design. They argue that intelligent data design is more
XGBoost. Systematic adjustments are made to parameters effective in improving analytical efficiency than developing
which incorporates mastering rate, maximum tree intensity, sophisticated new analytical techniques. The authors
and kind of boosting rounds. Through this improvement, the demonstrate this using a large European bank payment
set of guidelines plays higher and achieves the proper balance transaction data set and show that both feature design and
among precision and recall— metrics which can be essential request design significantly improve the performance of
to the identification of fraud [16] [17]. In the UPI Fraud analytical models. They also point out that simple analytical
Detection System, hyperparameter tuning is a critical step in methods such as logistic regression and classification trees can
pleasant-tuning the eXtreme Gradient Boosting (XGBoost) produce good results if the data are well designed. Although
the focus is on fraud in payment transactions, the techniques

925
Authorized licensed use limited to: ANURAG GROUP OF INSTITUTIONS. Downloaded on August 01,2024 at 09:05:56 UTC from IEEE Xplore. Restrictions apply.
 2024 2nd International Conference on Disruptive Technologies (ICDT)

discussed can be extended to other frauds in health care, [2] Fraud detection K-Nearest Easy to Didn’t
insurance or electronic commerce. with natural Neighbours implement work on
language Algorithm on small large
S. Alam [12], The XGBoost Classifier initially produced processing datasets datasets
the best results among the four classification algorithms used. [9] Data ROS and Easy to Loss of
To improve classifier performance, three different data Engineering for RUS implement valuable
balancing approaches were used, and ROS produced the best Fraud Detection on data
Fraud Detection imbalanced
results. RUS and the SMOTE method did not perform as well. dataset
With Random Over Sampling and the XGBoost Classifier,
accuracy, precision, recall, and F1 scores increased [12] Unbalanced XGBoost Easily Risk of
Credit Card handles overfitting
significantly. In the future, more effective approaches and Fraud Detection imbalanced if not
strategies should be explored to overcome the limitations of Data: A with appropriat
this study. Machine maximum ely tuned
Learning- accuracy
S. Mohanavalli [13], Data cleaning is a process that deals Oriented
with various data quality problems such as noise, outliers, Comparative
Study of
inconsistent data, duplicates and missing values. The focus is Balancing
on detecting and eliminating errors and inconsistencies during Techniques
the data collection phase. Traditional techniques mostly
[13] Survey of Pre- Logistic Easily Low
belong to this phase, but hybrid forms of error correction have processing Regression handle large accuracy
yet to be developed to provide noise-free samples for data Techniques for datasets as
mining and analysis. Different denoising techniques can lead Mining Big Data well as
to different databases and the median of the collected data can practical for
real-time
be used for homogeneous data sets, the range of values can be fraud
expanded for heterogeneous data sets. Future work includes detection
the investigation of hybrid methods for cleaning
[21] Principal PCA Easily May lack
homogeneous data and reducing noise in various Component reduces the clear
heterogeneous datasets using a voting system that combines Analysis noise and interpretab
multiple techniques. In addition, the impact of denoising redundancy ility
techniques on classification performance and mining [22] DeepSMOTE: SMOTE Preserves Risk of
performance should be investigated for large datasets. Fusing Deep information overfitting
Learning and by due to
M. Greenacre [21], A popular multivariate analytic SMOTE for generating synthetic
method for comprehending and exploring data is PCA Imbalanced Data synthetic cases
(Principal Component analysis). Its ability to extract cases in the
dataset
important information from complicated datasets has been
demonstrated by its application in a variety of areas. Large
datasets of all kinds may now be analyzed using PCA thanks
to recent developments, and this statistical technique should The above table 1 shows the comparison of the previous
continue to be improved by future breakthroughs. PCA is related research.
regarded as a fundamental tool in data science, along with its
III. PROPOSED METHODOLOGY
expansions and variants.
The below figure 3 shows the overall technique that we are
D. Dablain [22], To solve imbalanced data, a novel model implementing on our dataset. After dividing the dataset into
called DeepSMOTE combines deep learning and the well- training and test sets in which training data set contains 80%
known SMOTE technique. In order to balance the training set of the original data whereas the test set contains the remaining
and enable bias-free training of deep classifiers, it generates 20% of the original data. Then we will perform all these
fictitious instances. DeepSMOTE can create effective low- techniques on our training data set and in last we will use test
dimensional embeddings, interact with raw photos, and create data set to predict the transaction authenticity.
artificial images of excellent quality. According to
experimental research, DeepSMOTE works better than
alternative oversampling algorithms and is resilient to
different imbalance ratios. Subsequent efforts will focus on
augmenting DeepSMOTE with challenges at the class and
instance levels, refining the loss function, and expanding its
capabilities to accommodate dynamic class ratios and
additional data modalities like as text and graphs.

TABLE 1. COMPARATIVE STUDY OF PREVIOUS RESEARCH

Author Purposed Technolog Pros Cons
& Year Solution y
[1] Fraud detection Questionna Works well Didn’t Fig. 3. Procedure for detecting UPI fraud
model for ire- for old user work for
illegitimate Responded new user 1. Data Collection (Taking User Transaction Details):
transactions transaction
model Description:

926
Authorized licensed use limited to: ANURAG GROUP OF INSTITUTIONS. Downloaded on August 01,2024 at 09:05:56 UTC from IEEE Xplore. Restrictions apply.
 2024 2nd International Conference on Disruptive Technologies (ICDT)

Relevant information about user transactions is gathered Purpose:

in the first stage. This includes any information deemed
necessary for the fraud detection task, including timestamps, This stage establishes how the version have to be used in
transaction amounts, user information, and other facts. real situations. Accurately figuring out and flagging in all
likelihood fraudulent transactions is the reason, giving
Purpose: financial institutions or rate processors a device for
preference-making.
To train a strong fraud detection model, obtaining an
extensive dataset is essential. The dataset forms the basis for This flowchart describes a methodical procedure for
the process's further stages. detecting UPI fraud that begins with data collection, moves on
to correct imbalances, reduces dimensionality, trains a reliable
2. Pre-processing with SMOTE Algorithm: model, and concludes with decision-making based on the
Description: model's predictions. Every stage advances the model's
precision and efficacy in differentiating between authentic and
The dataset is pre-processed inside the 2nd degree using fraudulent transactions.
the Synthetic Minority Over-sampling Technique (SMOTE)
method. SMOTE is in particular designed to cope with IV. RESULT
information imbalances, particularly whilst the proportion of
When we fill all the mentioned transactions details we
legitimate transactions to fraudulent transactions is huge.
need to click on the “Click to Detect” button to check whether
Purpose: the transaction is Valid or Fraud transaction.
Biased models that carry out badly on minority training If the transaction is valid figure 4 will be shown as an
(fraudulent transactions) might possibly forestall end result output and if the transaction is fraud, then figure 5 will be
from imbalanced datasets. To balance the distribution of shown.
commands and enhance the version's fraud detection
capabilities, SMOTE creates artificial samples.
3. Feature Extraction with PCA:
Description:
Principal Component Analysis (PCA) is used in the 1/three
diploma to extract features. PCA is used to lessen the dataset's Fig. 4. Figure 1: Valid Transaction
dimensionality on the identical time as retaining important
data. This aids in overcoming the computational difficulties
added on by huge-scale statistics.
Purpose:
By using PCA to lessen dimensionality, the dataset
becomes extra practicable for system learning algorithms. The
motive of extracted abilties is to extract the most critical
information, which improves model performance.
Fig. 5. Fraud Transaction
4. Model Training with XGBoost:
Description: In figure 6 we can see the comparison of multiple
algorithms in which XGBoost outperforms all the algorithms.
The fourth segment includes schooling a system learning
model, XGBoost, the use of the pre-processed and function- >ŽŐŝƐƚŝĐZĞŐƌĞƐƐŝŽŶ ϳϴ͘ϵ
extracted dataset. XGBoost is a famous ensemble gaining ϴϬ͘Ϯ
knowledge of set of policies that excels at dealing with ^ƵƉƉŽƌƚsĞĐƚŽƌDĂĐŚŝŶĞ ϴϭ͘ϱ
ϴϯ͘ϰ
complex, non-linear relationships in statistics because of its ĞĐŝƐŝŽŶdƌĞĞ ϴϴ͘ϳ
tremendous traditional performance and performance. ϵϮ͘ϲ
y'ŽŽƐƚ ϵϴ͘Ϯ
Purpose: Accuracy
XGBoost is selected because of its versatility in managing
unbalanced datasets, noise resilience, and capability to Fig. 6. Accuracy Comparison
perceive complex styles inside the records. Based on the skills
which have been processed and altered, the version is skilled The below table 2 shows the comparison of all algorithms.
to distinguish amongst real and fraudulent transactions.
TABLE 2. COMPARISON OF ACCURACY
5. Decision (Real or Fake Transaction):
Algorithm Name Accuracy Score (%)
Description:
XGBoost 98.2
Making predictions on new, unseen transactions the use of Random Forest 92.6
the informed XGBoost version is the final stage. The version
uses the styles it has learnt in some unspecified time in the Decision Tree 88.7
future of training to categorize every transaction as both real K-Nearest Neighbors 83.4
or fraudulent.

927
Authorized licensed use limited to: ANURAG GROUP OF INSTITUTIONS. Downloaded on August 01,2024 at 09:05:56 UTC from IEEE Xplore. Restrictions apply.
 2024 2nd International Conference on Disruptive Technologies (ICDT)

Support Vector Machine 81.5 for instant payments. Complex Systems Informatics and Modeling
Quarterly, (20), 72-88. https://doi.org/10.7250/csimq.2019-20.04
Naive Bayes 80.2
[9] Baesens, B., Höppner, S., & Verdonck, T. (2021). Data engineering for
Logistic Regression 78.9 fraud detection. Decision Support Systems, 150,
113492.https://doi.org/10.1109/ICICV50876.2021.9388431
[10] Carminati, M., Baggio, A., Maggi, F., Spagnolini, U., & Zanero, S.
(2018). FraudBuster: temporal analysis and detection of advanced
V. CONCLUSION financial frauds. In Detection of Intrusions and Malware, and
Vulnerability Assessment: 15th International Conference, DIMVA
This UPI fraud detection project has applied innovative 2018, Saclay, France, June 28–29, 2018, Proceedings 15 (pp. 211-233).
methodologies to enhance the robustness and efficiency of our Springer International,Publishing.
fraud detection model. By addressing the imbalanced nature https://doi.org/10.1016/j.procs.2023.01.231
of the dataset with the Synthetic Minority Over-sampling [11] Rastogi, S., Sharma, A., Panse, C., & Bhimavarapu, V. M. (2021).
Technique (SMOTE), we have successfully balanced the Unified Payment Interface (UPI): A digital innovation and its impact
representation of fraudulent transactions, boosting the model's on financial inclusion and economic development. Universal Journal of
Accounting and Finance, 9(3),518-530.
ability to recognize subtle patterns indicative of fraudulent https://doi.org/10.1109/ICCCSP.2017.7944072
operations. [12] Gupta, P., Varshney, A., Khan, M. R., Ahmed, R., Shuaib, M., & Alam,
The application of Principal Component Analysis (PCA) S. (2023). Unbalanced Credit Card Fraud Detection Data: A Machine
Learning-Oriented Comparative Study of Balancing Techniques.
for feature extraction has contributed to the reduction of Procedia Computer-Science,218,2575-2584.
dimensionality, capturing the most salient aspects of the data https://doi.org/10.3390/bdcc7020093
and mitigating issues related to multicollinearity. This [13] Hariharakrishnan, J., Mohanavalli, S., & Kumar, K. S. (2017, January).
simplified the computing procedure and made it easier to Survey of pre-processing techniques for mining big data. In 2017
understand the underlying fraud tendencies in the international conference on computer, communication and signal
representation. processing (ICCCSP) (pp. 1-5). IEEE.
https://doi.org/10.33050/atm.v5i2.1593
Our choice of XGBoost as the training model has been [14] Bhattacharyya, S., Jha, S., Tharakunnel, K., & Westland, J. C. (2011).
effective, exploiting its capabilities in managing skewed data, Data mining for credit card fraud: A comparative study. Decision
high predicted accuracy, and robustness. XGBoost's capacity support systems, 50(3), 602-613.
https://doi.org/10.1016/j.softx.2019.100341
to capture intricate relationships within the data has
[15] Branco, B., Abreu, P., Gomes, A. S., Almeida, M. S., Ascensão, J. T.,
considerably boosted the model's overall effectiveness in & Bizarro, P. (2020, August). Interleaved sequence rnns for fraud
recognizing probable cases of UPI fraud. detection. In Proceedings of the 26th ACM SIGKDD international
conference on knowledge discovery & data mining (pp. 3101 3109).
For future scope we can reduce the number of transactions https://doi.org/10.33369/jakuntansi.12.3.181-201
details [23] [24] that is needed to detect the transaction [16] Zhu, B., Gao, Z., Zhao, J., & vanden Broucke, S. K. (2019). IRIC: An
authenticity whereas as well as we can track trends in user R library for binary imbalanced classification.-SoftwareX,10,100341.
interactions with the UPI platform by integrating analytics on https://doi.org/10.13189/ujaf.2021.090326
user behavior. Behavior anomalies may be used as early [17] Dileep, M. R., Navaneeth, A. V., & Abhishek, M. (2021, February). A
warning signs of possible fraud. novel approach for credit card fraud detection using decision tree and
random forest algorithms. In 2021 Third International Conference on
REFERENCES Intelligent Communication Technologies and Virtual Mobile Networks
(ICICV) (pp. 1025-1028). IEEE.
[1] Adekunle, I. M., & Ozoh, P. (2023). Fraud detection model for https://doi.org/10.33050/atm.v5i2.1593
illegitimate transactions. Kabale University Interdisciplinary Research
Journal, 2(2), 21-37 https://doi.org/10.1016/j.future.2015.01.001 [18] Ahmed, M., Mahmood, A. N., & Islam, M. R. (2016). A survey of
anomaly detection techniques in financial domain. Future Generation
[2] Boulieris, P., Pavlopoulos, J., Xenos, A., & Vassalos, V. (2023). Fraud Computer Systems, 55, 278-288.
detection with natural language processing. Machine Learning, 1 22. https://doi.org/10.1109/ICCCSP.2017.7944072
https://doi.org/10.24321/2394.6539.202012
[19] Lavadkar, M. A., Thorat, P. K., Kasliwal, A. R., Gadekar, J. S., &
[3] Mytnyk, B., Tkachyk, O., Shakhovska, N., Fedushko, S., & Syerov, Y. Deshmukh, D. P. Fingerprint Biometric Based Online Cashless
(2023). Application of Artificial Intelligence for Fraudulent Banking Payment System. IOSR Journal of Computer Engineering (IOSR-JCE),
Operations Recognition. Big Data and Cognitive-Computing,7(2),93. e-ISSN, 2278-0661. https://doi.org/10.1016/j.softx.2019.100341
https://doi.org/10.1016/j.dss.2010.08.008
[20] Purnama, S., Bangun, C. S., & Faaroek, S. A. (2021). The Effect of
[4] Ridwan, R., Abdullah, S., & Yusmita, F. (2022). IMPLEMENTATION Transaction Experience Using Digital Wallets on User Satisfaction in
OF CASHLESS POLICY STRATEGIES TO MINIMIZE FRAUD IN Millennial Generation. Aptisi Transactions on Management (ATM),
THE GOVERNMENTSECTOR: SYSTEMIC REVIEW. Jurnal 5(2), 161-168. https://doi.org/10.24321/2394.6539.202012
Akuntansi, 12(3), 181-201. https://doi.org/10.1007/s10994-023-
06354-5 [21] Greenacre, M., Groenen, P. J., Hastie, T., d’Enza, A. I., Markos, A., &
Tuzhilina, E. (2022). Principal component analysis. Nature Reviews
[5] Chang, V., Di Stefano, A., Sun, Z., & Fortino, G. (2022). Digital Methods Primers, 2(1), 100.
payment fraud detection methods in digital ages and Industry 4.0. https://doi.org/10.13189/ujaf.2021.090326
Computers and Electrical Engineering,100, 107734.
https://doi.org/10.1145/3394486.3403361 [22] Dablain, D., Krawczyk, B., & Chawla, N. V. (2022). DeepSMOTE:
Fusing deep learning and SMOTE for imbalanced data. IEEE
[6] Bandyopadhyay, S. K., & Dutta, S. (2020). Detection of fraud Transactions on Neural Networks and Learning-Systems.
transactions using recurrent neural network during COVID-19: fraud https://doi.org/10.1109/TNNLS.2021.3136503
transaction during COVID-19. Journal of Advanced Research in
Medical Science & Technology (ISSN: 2394-6539),7(3),16-21. [23] Rani,R., Yogi, K.K., Yadav, S.P.,. Detection of Cloned Attacks in
https://doi.org/10.1016/j.compeleceng.2022.107734 Connecting Media using Bernoulli RBM_RF Classifier (BRRC).
Multimed Tools Appl (2024).
[7] Manocha, S., Kejriwal, R., & Upadhyaya, D. A. (2019, September).
The impact of demonetization on digital payment transactions: a [24] Kanaujia, V.K., Kumar, A., Yadav, S.P., Advancements in Automatic
statistical study. In Proceedings ofInternational Conference on Kidney Segmentation using Deep Learning Frameworks and
Advancements in Computing & Management (ICACM). Volumetric Segmentation Techniques for CT Imaging: A Review,
https://doi.org/10.1109/TNNLS.2021.3136503 Arch Computat Methods Eng 29, 1753–1770 (2024)
[8] Diadiushkin, A., Sandkuhl, K., & Maiatin, A. (2019). Fraud detection
in payments transactions: Overview of existing approaches and usage

928
Authorized licensed use limited to: ANURAG GROUP OF INSTITUTIONS. Downloaded on August 01,2024 at 09:05:56 UTC from IEEE Xplore. Restrictions apply.