Predictive Analytics for Digital Transactions

Sasikala C., Anil Kumar Bandi, Ambica Cheluru, Aswartha Reddy Settipi,

Durga Bhavani Vanka and Tarun Kumar Reddy Peram

Department of Computer Science and Engineering, Srinivasa Ramanujan Institute of Technology (SRIT), Anantapur,

Andhra Pradesh, India

Keywords: Choice Trees, Irregular Woodlands, Angle Assistance, Protection Safeguarding Models, Models of

Straightforwardness Include Deep Neural Networks (DNN), Recurrent Neural Networks (RNN), Federated

Learning (FL), Financial Fraud Detection, Explainable Artificial Intelligence (XAI), AI Algorithms.

Abstract: In the domain of monetary misrepresentation location, accomplishing harmony among straightforwardness

and security is basic. Conventional methodologies frequently neglect to give both elevated degrees of

exactness and clear, justifiable clarifications for navigation, all while safeguarding delicate information. They

attempt to uncover the solutions that United Learning (FL) and Reasonable Computer-based Intelligence

(XAI) provide on these particular challenges. Financial institutions utilizing quantitative predictions must

deeply trust their models, and XAI provides models for them. However, Unified Learning considers the

construction of AI models to a collection of dispersed data sources where sensitive financial data is properly

siloed. In detail, this study looks at how different algorithms, which include Deep Neural Networks (DNN),

Recurrent Neural Networks (RNN), Decision Trees, Random Forests, and Support Vector Machines, are

utilized in fraud detection tasks and how these algorithms are integrated into the multi-task learning

framework. In addition, the study investigates fusion methods for models such as Stochastic Gradient Descent

(SGD) and its variants. This study investigates how financial institutions could enhance fraud detection

systems while ensuring transparency, confidentiality, and compliance with data protection laws by integrating

the best of both XAI and FL worlds.

1 INTRODUCTION

This article examines the combination of Explainable

AI (XAI) and Federated Learning (FL) with the goal

of increasing fraud detection capabilities within

financial institutions while enhancing transparency,

regulatory compliance, and data security. AI-enabled

fraud detection provides high accuracy, however, as

explained previously, there is a significant lack of

explain ability, which complicates matters within the

extremely controlled financial sector. To foster trust

and compliance with regulations like GDPR, XAI

ensures that decision-making by AI systems is

more explainable and transparent. On the other

hand, FL enables the development of AI models

on data held in silos, thereby preserving privacy and

minimizing the risk of security breaches. Established

financial institutions can now build effective systems

for accurate fraud detection without compromising

security, explain ability, or precision by integrating

XAI and FL. The models selected for the study

include decision trees, recurrent neural networks

(RNN), deep neural networks (DNN), and gradient

boosting models with a focus on the FL framework

for AI. The focus is also on optimization methods

such as stochastic gradient descent (SGD) to make

them more efficient. Lastly, the paper proposes a

framework for responsible use of AI in financial

fraud detection by achieving a balance of data

privacy, transparency, and overall system

performance.

2 THE ROLE OF THIS WEB

APPLICATION

The primary purpose of this study is to assess the

effectiveness of a novel combined approach using

Federated Learning (FL) and Explainable AI (XAI)

in payment fraud detection systems. There are

existing techniques for fraud detection and

prevention that work, but they tend to be centralized

C., S., Bandi, A. K., Cheluru, A., Settipi, A. R., Vanka, D. B. and Peram, T. K. R.

Predictive Analytics for Digital Transactions.

DOI: 10.5220/0013922500004919

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies (ICRDICCT‘25 2025) - Volume 5, pages

56-61

ISBN: 978-989-758-777-1

and obfuscated, which can lead to challenges with

privacy and compliance laws. FL does allow for data

processing to be conducted in a more decentralized

manner, but XAI claims to provide trustable AI

conclusions. The focus of this research is on the

comparison of different AI models, including Deep

Learning Neural Networks (DNNs), Recurrent

Neural Networks (RNNs), Decision Trees, Random

Forest, and Gradient Boosting, which will all be

executed under FL framework. In order to improve

the performance of the models, additional parameters

such as Stochastic Gradient Descent (SGD) will also

be utilized. It is the objective of this work to also

illustrate an E2E framework for AI augmented

payment fraud detection featuring accuracy,

transparency, data privacy, and regulatory

compliance.

3 LITERATURE REVIEW

As scientists strive to enhance model simplicity,

security, and predictive accuracy, the combination of

Federated Learning (FL) and Explainable Artificial

Intelligence (XAI) has recently gained considerable

attention, especially in the realm of financial fraud

detection. Traditional AI methods have primarily

depended on centralized models for identifying

misrepresentation, where sensitive financial data is

stored and managed on centralized servers. However,

these methods face challenges related to regulatory

compliance and security, particularly with stringent

data protection laws like the GDPR. A promising

solution to these challenges is Federated Learning,

which facilitates the collaborative development of

models without sharing raw data among participants.

FL enables various financial institutions or

organizations to collaboratively build AI models

using their local datasets while keeping their most

sensitive information private, thus addressing

security concerns while leveraging extensive data

resources. Since the variety of value-based designs

employed by various organizations is essential in

spotting unusual activities, this widely recognized

learning paradigm is especially useful in detecting

financial extortion. Simultaneously, the test lies in

guaranteeing that such models are interpretable and

straightforward, as monetary (Balcıoğlu, Y. S)

foundations should give clarifications to their

mechanized choices, particularly when these choices

can have critical legitimate and monetary outcomes.

The requirement for interpretability in

misrepresentation location models has prompted a

developing interest in XAI. Conventional AI

calculations, especially profound learning models,

are frequently seen as "secret elements" because their

(Bodker A et al., 2022) dynamic cycles are not

effectively justifiable by people. This absence of

straightforwardness creates difficulties in managed

areas like money, where it isn't simply vital to make

exact expectations but additionally to give clear

defenses to those forecasts. ( Demertzis et al., 2022

XAI expects to resolve this issue by creating

procedures and models that permit leaders to

comprehend and believe the expectations made by

simulated intelligence frameworks. Different

procedures have been proposed to upgrade the

interpretability of mind- boggling models, for

example, highlight significance examination, nearby

clarification techniques like LIME (Neighborhood

Interpretable (Guo et al., 2024) Model rationalist

Clarifications), and SHAP (SHapley Added

substance Clarifications). These techniques offer a

method for making sense of individual expectations,

making it clearer why a specific exchange was hailed

as fake or not. By giving partners straightforward bits

of knowledge into the dynamic cycle, XAI not only

upgrades trust in computerized misrepresentation

identification frameworks (Hasan et al., 2024) yet in

addition helps meet administrative prerequisites in

regard to responsibility and decency in a computer-

based intelligence-driven direction. As far as

algorithmic commitments to misrepresentation

location, an extensive variety of AI models have been

investigated. Profound Brain Organizations (DNNs)

have been widely utilized because of their capacity to

catch complex examples and learn progressive

portrayals of information. These models (Koetsier et

al., 2022) have shown great outcomes in

distinguishing misrepresentation, especially in

conditions where the false way of behaving is

unobtrusive or advances over the long run.

Nonetheless, (Kollu et al., 2023) DNNs additionally

face difficulties connected with interpretability, as

they can be exceptionally perplexing and hard to

make sense of.

Repetitive Brain Organizations (RNNs),

especially Lengthy Transient Memory (LSTM)

organizations, have additionally been utilized in

misrepresentation (Lakhan etal., 2023) recognition

undertakings that include consecutive information,

like exchange chronicles. RNNs are successful in

demonstrating fleeting conditions, which is basic in

monetary extortion location, as deceitful exercises

frequently unfurl after some time and show

consecutive examples.

While RNNs offer benefits as far as transient

examination, they likewise experience the ill effects

Predictive Analytics for Digital Transactions

of interpretability issues, prompting a developing

interest in coordinating XAI procedures with RNN-

based models to offer more straightforward

misrepresentation discovery arrangements. Other

conventional AI procedures, for example, Choice

Trees, (Mothukuri et al., 2021) Arbitrary Woods, and

Angle Supporting techniques, have additionally been

generally utilized in misrepresentation locations.

Choice Trees are well known because of their

intrinsic interpretability; their treelike design

considers simple perceptions of choice ways. Be that

as it may, their exhibition can corrupt while taking

care of enormous, high-layered datasets, prompting

the utilization of group techniques like Irregular

Timberlands and Inclination (Raval et al., 2023)

Helping Machines (GBM). These techniques, while

further developing precision, present intricacy,

making them harder to decipher. In any case, there

have been propels in posthoc interpretability

techniques for troupe models, for example,

highlighting significance scores and proxy models,

which have helped address the compromise between

model exactness and logic. In addition, the utilization

of group models has been displayed to further

develop recognition (Sai et al., 2023) rates by joining

the qualities of different models and lessening the

gamble of overfitting.

4 IMPLEMENTATIONS

4.1 Flow Chart

A flowchart is a visual representation of the processes

occurring within a system or project. It illustrates the

various steps involved. The flowchart starts and

concludes at the terminal points, which are depicted

using oval shapes. Decision-making steps are

represented by diamond shapes. Rectangular boxes

indicate the processes that occur on the website.

Processes that are adjacent to each other are

sequential. Figure 1 Shows the Flow chart of

application.

4.2 Database Connectivity

To store and access data from the server we require a

database connection and we implanted below code

snippet below provide the connection. This code

creates a new mysqli object, specifying the database

server, username, password, and database name. The

connection is then verified, and an error message is

displayed if the connection fails.

Figure 1: Flow Chart of Application.

4.3 Fetching Data

To fetch data from the database we implemented the

following code snippet to get the data and load it in the

frontend pages. This query retrieves the user's data

from the users table, based on their unique user_id.

The resulting data is then processed and displayed

within the application. The shows Figure 2 Register

page.

Figure 2: Register Page.

4.4 Front end

The frontend of our web application was built using

HTML, CSS, JavaScript, and jQuery. HTML

provided the structural foundation, while CSS was

used for styling and layout. JavaScript and jQuery

were utilized for dynamic client-side functionality,

enhancing user interaction and experience. Our

frontend design aimed to provide an intuitive and

user- friendly interface, allowing users to seamlessly

navigate and utilize the application's features. The

Figure 3 shows Login Page.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

Figure 3: Login Page.

4.5 User Credentials Validation

This is an example code snippet of JavaScript used in

our application for front-end. This code snippet

verifies credentials and logs in the students and

faculty by sending a request to the database. It

verifies the user credentials with in the browser

before sending to the server for backend verification.

The shows Figure 4 Home page.

Figure 4: Home Page.

5 RESULTS

The interfaces of this progressive web application are

shown below Figure 5 and 6.

Figure 5: Model Training.

After selecting a certain algorithm, we calculate their

accuracies based on the result of the result of the

accuracies we choose which algorithm is best for

fraud detection. The process of choosing the

appropriate algorithm for fraud detection is illustrated

in Figure 6, while the final output of the system is

demonstrated on the prediction page shown in Figure

Figure 6: Model Selection.

Figure 7: Prediction Page.

The below analytics shows Figure 8 and 9 the

performance of the web application. In case the given

values are legitimate then it shows the No fraud

transaction.

Figure 8: Result 1.

In case the given values aren’t legitimate then it shows

the fraud transaction.

Predictive Analytics for Digital Transactions

Figure 9: Result 2.

6 CONCLUSIONS

By ensuring data privacy through decentralized

model training without exposing individual data,

federated learning (FL) is enhanced by explainable

AI (XAI), which makes AI-driven decisions more

interpretable. This collaboration guarantees the

development of reliable, interpretable, and legally

compliant fraud detection systems. The study

illustrates that while deep learning models like deep

neural networks (DNNs) and recurrent neural

networks (RNNs) offer high precision, they often

lack transparency. In contrast, traditional models

such as decision trees and random forests provide

explain ability but may fall short in precision. By

integrating these models into a federated learning

framework, a hybrid approach can achieve a balance

of explain ability, precision, and data protection.

These advancements offer financial institutions a

roadmap for implementing effective and

regulatory-compliant fraud detection technologies,

adhering to standards like the CCPA and GDPR.

7 ACKNOWLEDGMENT

We extend our heartfelt thanks to Dr. C. Sasikala, an

assistant professor of computer science and engineering at

Srinivasa Ramanujan Institute of Technology, for his

invaluable guidance and support in making this research

a success.

REFERENCES

Ahmed, A. A., & Alabi, O. O. (2024). Secure and Scalable

Blockchain-Based Federated Learning for

Cryptocurrency Fraud Detection: A Systematic

Review. IEEE Access, 12, 102219–

102241.https://doi.org/10.1109/ACCESS.2024.342920

Ali, S., Li, Q., & Yousafzai, A. (2024). Blockchain and

federated learning-based intrusion detection

approaches for edge-enabled industrial IoT networks: a

survey. Ad Hoc Networks, 152, 103320.

https://doi.org/10.1016/J.ADHOC.2023.103320

Attanayaka, D. (2022). A novel anomaly detection

mechanism for Open radio access networks with Peer-

to- Peer Federated Learning. Laturi.Oulu.Fi.

https://oulurepo.oulu.fi/handle/10024/21293

Balcıoğlu, Y. S. (1 C.E.). Revolutionizing Risk

Management AI and ML Innovations in Financial

Stability and Fraud Detection.

Https://Services.IgiGlobal.Com/Resolvedoi/Resolve.A

spx?Doi=10.4018/979-8-3693-4382- Ch006,109–138.

https://doi.org/10.4018/979-8-3693-4382-1.CH006

Bodker, A., Connolly, P., Sing, O., Hutchins, B., Townsley,

M., & Drew, J. (2022). Card-not-present fraud: using

crime scripts to inform crime prevention initiatives.

Security Journal, 36(4), 1.

https://doi.org/10.1057/S41284-022-00359-W

Demertzis, K., Iliadis, L., Kikiras, P., & Pimenidis, E.

(2022). An explainable semi-personalized federated

learning model. Integrated Computer-Aided

Engineering, 29(4), 335–350.

https://doi.org/10.3233/ICA-220683

Guo, W., & Jiang, P. (2024). Weakly Supervised anomaly

detection with privacy preservation under a Bi- Level

Federated learning framework. Expert Systems with

Applications, 254, 124450.

https://doi.org/10.1016/J.ESWA.2024.124450

Hasan, M., Rahman, M. S., Janicke, H., & Sarker, I. H.

(2024). Detecting anomalies in blockchain transactions

using machine learning classifiers and explainability

analysis. Blockchain: Research and Applications, 5(3),

100207. https://doi.org/10.1016/J.BCRA.2024.100207

Koetsier, C., Fiosina, J., Gremmel, J. N., Müller, J. P.,

Woisetschläger, D. M., & Sester, M. (2022). Detection

of anomalous vehicle trajectories using federated

learning. ISPRS Open Journal of Photogrammetry and

Remote Sensing, 4, 100013.

https://doi.org/10.1016/J.OPHOTO.2022.100013

Kollu, V. N., Janarthanan, V., Karupusamy, M., &

Ramachandran, M. (2023). Cloud-Based Smart

Contract Analysis in FinTech Using IoT-Integrated

Federated Learning in Intrusion Detection. Data 2023,

Vol. 8, Page 83, 8(5), 83.

https://doi.org/10.3390/DATA8050083

Lakhan, A., Mohammed, M. A., Nedoma, J., Martinek, R.,

Tiwari, P., Vidyarthi, A., Alkhayyat, A., & Wang, W.

(2023). Federated-Learning Based Privacy

Preservation and Fraud-Enabled Blockchain IoMT

System for Healthcare. IEEE Journal of Biomedical and

Health Informatics, 27(2), 664–672.

https://doi.org/10.1109/JBHI.2022.3165945

Marry, P., Mounika, Y., Nanditha, S., Shiva, R., &

Saikishore, R. (2024). Federated Learning-Driven

Decentralized Intelligence for Explainable Anomaly

Detection in Industrial Operations. 2nd International

Conference on Sustainable Computing and Smart

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

Systems, ICSCSS 2024 - Proceedings, 874–880.

https://doi.org/10.1109/ICSCSS60660.2024.10625289

Mothukuri, V., Parizi, R. M., Pouriyeh, S., Huang, Y.,

Dehghantanha, A., & Srivastava, G. (2021). A

survey on security and privacy of federated learning.

Future Generation Computer Systems, 115, 619640.

https://doi.org/10.1016/J.FUTURE.2020.10.007

Raval, J., Bhattacharya, P., Jadav, N. K., Tanwar, S.,

Sharma, G., Bokoro, P. N., Elmorsy, M., Tolba, A., &

Raboaca, M. S. (2023). RaKShA: A Trusted

Explainable LSTM Model to Classify Fraud Patterns on

Credit Card Transactions. Mathematics 2023, Vol.

11, Page 1901, 11(8),1901. https://doi.org/10.3390/M

ATH11081901

Sai, C. V., Das, D., Elmitwally, N., Elezaj, O., & Islam, M.

B. (2023) Explainable Ai-Driven Financial Transaction

Fraud Detection Using Machine

Learning and Deep Neural Networks. https://doi.org/1

0.2139/SSRN.4439980

Predictive Analytics for Digital Transactions