FORM NO.

F/ TL / 024
Rev.00 Date 20.03.2020

FRAUD DETECTION USING DEEP

LEARNING
MINI PROJECT REPORT

submitted in partial fulfillment of the

requirements for the award of the degree in

BACHELOR OF TECHNOLOGY
IN
COMPUTER SCIENCE AND ENGINEERING - AI

by

YASWANTH.M (211211101370)
U.MANOJ KUMAR (211211101334)

DEPARTMENT OF

COMPUTER SCIENCE AND ENGINEERING-AI

MAY 2024
 DEPARTMENT OF
COMPUTER SCIENCE AND ENGINEERING-AI

BONAFIDE CERTIFICATE

This is to certify that this Project Report is the bonafide work of

Mr. YASWANTH.M Reg. No 211211101370,

Mr. U. MANOJ KUMAR Reg. No 211211101334 who carried out the

project entitled “FRAUD DETECTION USING DEEP LEARNING”

under our supervision from January 2024 to May 2024

Mini project Mini Project HOD

Coordinator 1 coordinator 2

Mrs.C.SUBA LAKSHMI Dr.M.UMA MAHESWAR DR S.GEETHA

Assistant Professor Assistant Professor Hod CSE
CSE Department CSE Department CSE Department
Dr.M.G.R.Educational Dr.M.G.R.Educational Dr.M.G.R.Educational
and Research Institute and Research Institute and Research Institute
Deemed to be University Deemed to be University Deemed to be
University

Submitted for Viva Voce Examination held on

Internal Examiner External Examiner

 DECLARATION

We YASWANTH.M (211211101370), U. MANOJ KUMAR (211211101334) hereby

declare that the Project Report entitled “FRAUD DETECTION USING DEEP

LEARNING” is done by us under the guidance of Dr./Prof./Mr./Ms C. SUBA LAKSHMI

is submitted in partial fulfillment of the requirements for the award of the degree in

BACHELOR OF TECHNOLOGY IN COMPUTER SCIENCE AND ENGINEERING - AI.

1.

2.

DATE:

PLACE: SIGNATURE OF THE CANDIDATE

 ACKNOWLEDGEMENT

We would like to thank our beloved Chancellor Thiru. Dr.A.C. Shanmugam, B.A.,

B.L., President Er. A.C.S. Arunkumar, B.Tech., and Secretary Thiru A.Ravikumar

for all the encouragement and support extended to us during the tenure of this project

and also our years of studies in this wonderful University.

We express our heartfelt thanks to our Vice Chancellor Dr. S.Geethalakshmi in

providing all the support of my Project.

We express our heartfelt thanks to our Head of the Department, Prof. Dr. S.

Geetha, who has been actively involved and very influential from the start till the

completion of our Project.

Our sincere thanks to our Internal Guides Mrs. C. SUBA LAKSHMI & Dr. M. UMA

MAHESWAR and Project guide C. SUBA LAKSHMI for their continuous guidance

and encouragement throughout this work, which has made the project a success.

We would also like to thank all the teaching and non teaching staff of Computer

Science and Engineering department, for their constant support and the

encouragement given to us while we went about to achieving our project goals.

 TABLE OF CONTENT

CHAPTER NO TITLE PAGE NO

1 ABSTRACT 1

2 2.1 INTRODUCTION 2

2.2 PROJECT OVERVIEW

2.3 AIM OF THE PROJECT

2.4SCOPE OF THE PROJECT

3 LITERATURE REVIEW 5

4 ANALYSIS 6

4.1 EXSISTING SYSTEM

4.2 PROPOSED SYSTEM

5 FLOW CHART 8

6 REQUIREMENT ANALYSIS 9

7 IMPLEMENTATION 10

7.1 CODING

8 RESULT 14

9 9.1 CONCLUSION 16

9.2 FUTURE WORKS

10 REFERNCES 20
 CHAPTER 1
ABSTRACT

Fraudulent activities pose a significant threat to the integrity of financial

systems, necessitating advanced detection mechanisms. In this project, we
propose the development of a sophisticated fraud detection system utilizing
cutting-edge deep learning methodologies. The system is engineered to
meticulously analyze customer behavior in real-time, ensuring swift
identification of suspicious transactions.Central to our approach is the
utilization of a meticulously curated dataset comprising both fraudulent and
non-fraudulent transactions. This dataset serves as the foundation for
comprehensive training, enabling the system to discern subtle patterns
indicative of fraudulent behavior. Leveraging the power of deep learning
algorithms, the system undergoes rigorous training to learn and adapt to
evolving fraudulent strategies.To assess the efficacy of our system, evaluation
is conducted on a holdout dataset, providing a robust measure of its accuracy
and effectiveness. By meticulously analyzing its performance metrics, we
ascertain the system's ability to accurately detect fraudulent activities while
minimizing false positives.Our overarching objective is to create a highly
efficient and reliable fraud detection solution, capable of providing robust
protection to financial institutions and their clientele. Through the integration
of cutting-edge deep learning techniques, we aim to bolster security measures
within the financial sector, safeguarding the interests of stakeholders against
fraudulent endeavors.

1
 CHAPTER 2

2.1 INTRODUCTION

In today's rapidly evolving financial landscape, the rise of increasingly

sophisticated fraudulent activities presents a significant challenge for financial
institutions worldwide. Detecting and preventing fraud in real-time has
become imperative to safeguarding the interests of both businesses and
consumers. However, traditional fraud detection methods, relying on static
rules and statistical models, often struggle to keep pace with the dynamic
nature of emerging fraud patterns.

In response to these challenges, our project endeavors to pioneer an

innovative approach to fraud detection by harnessing the power of deep
learning methodologies. By leveraging advanced deep learning algorithms,
we aim to develop a robust system capable of analyzing customer behavior
in real-time and promptly detecting fraudulent transactions. Through
comprehensive training on a labeled dataset containing a diverse range of
both fraudulent and non-fraudulent transactions, our system seeks to learn
intricate patterns indicative of fraudulent activities. The subsequent evaluation
of the system's performance on a holdout dataset will offer valuable insights
into its accuracy and effectiveness.

This introduction serves as a prelude to our project's objectives, highlighting

the pressing need for cutting-edge solutions to combat financial fraud
effectively. By embracing the transformative potential of deep learning, our
ultimate goal is to contribute to the advancement of fraud detection
technology, thereby bolstering security measures within the financial sector
and fostering greater trust among stakeholders.

2
2.2 PROJECT OVERVIEW

The project aims to pioneer an innovative approach to fraud detection by

leveraging deep learning methodologies in the rapidly evolving financial
landscape. Traditional fraud detection methods often struggle to keep pace
with the dynamic nature of emerging fraud patterns, necessitating advanced
solutions. Our system will utilize advanced deep learning algorithms to
analyze customer behavior in real-time, facilitating prompt detection of
fraudulent transactions. Through comprehensive training on a labeled dataset
containing diverse transactions, the system will learn intricate patterns
indicative of fraudulent activities. Evaluation on a holdout dataset will provide
insights into the system's accuracy and effectiveness. Ultimately, the project
seeks to contribute to the advancement of fraud detection technology,
enhancing security measures within the financial sector and fostering greater
trust among stakeholders.

2.3 AIM OF THE PROJECT

The aim of the project is to pioneer an innovative approach to fraud detection

in the financial sector by leveraging deep learning methodologies. Traditional
methods of fraud detection often rely on static rules and statistical models,
which may struggle to adapt to the constantly evolving tactics employed by
fraudsters. Therefore, our objective is to develop a sophisticated fraud
detection system that can analyze real-time customer behavior and promptly
identify fraudulent transactions.

By harnessing the power of deep learning algorithms, we aim to create a

system capable of learning complex patterns indicative of fraudulent activities.
This entails training the system on a meticulously labeled dataset comprising
a wide range of both fraudulent and non-fraudulent transactions. Through this
comprehensive training process, the system will be equipped to detect subtle
anomalies and deviations from normal behavior that may signal potential
fraud.

3
 Furthermore, the project aims to conduct thorough evaluation and testing of
the developed system on a holdout dataset. This evaluation phase is crucial
for assessing the system's accuracy, effectiveness, and robustness in real-
world scenarios. By rigorously evaluating the system's performance, we can
identify areas for improvement and fine-tune its algorithms to enhance its
overall efficacy in detecting fraud.

2.4 SCOPE OF THE PROJECT

The scope of the project involves developing a sophisticated fraud detection

system using deep learning methodologies within the financial sector. This
encompasses data collection, preprocessing, and labeling, followed by the
selection and implementation of appropriate deep learning algorithms. The
project includes training and evaluating the models using diverse datasets to
ensure accurate detection of fraudulent activities. Real-time monitoring and
detection capabilities will be integrated into existing financial systems, with
continuous performance monitoring and optimization to maintain effectiveness
over time. Deployment and integration into production environments will be
managed, accompanied by comprehensive documentation and knowledge
transfer to relevant stakeholders. Overall, the project aims to enhance security
measures and mitigate financial risks through advanced fraud detection
techniques.

4
 CHAPTER 3

LITERATURE REVIEW

The literature reviewed encompasses a range of studies aimed at enhancing

credit card fraud detection through the application of machine learning and
deep learning techniques. Raghavan and El Gayar (2019) proposed
methodologies for fraud detection at the 2019 ICCIKE conference, likely
focusing on the integration of these techniques into real-world financial
systems. Pumsirirat and Liu (2018), Pandey (2017), and Pillai et al. (2018)
explored deep learning approaches such as auto-encoder and RBM
architectures for improved fraud detection accuracy. Similarly, Babu and
Pratap (2020) investigated novel deep learning techniques to enhance
detection capabilities. Bahnsen et al. (2016) and Bhattacharyya et al. (2011)
discussed feature engineering and the combination of unsupervised and
supervised learning methods, while Carcillo et al. (2021) provided further
insights into this integration for enhanced detection accuracy. Additionally,
Shenvi et al. (2019) and Beigi and Amin Naseri (2020) contributed to the field
with studies on deep learning methodologies and the integration of data
mining and statistical methods, respectively, likely focusing on novel
techniques or the optimization of existing approaches. Collectively, these
studies highlight the significance of leveraging advanced machine learning
and deep learning techniques to mitigate credit card fraud, contributing to the
ongoing efforts to bolster security measures within the financial sector.

5
 CHAPTER 4

ANALYSIS

4.1 EXISTING SYSTEM

Existing fraud detection systems typically employ rule-based or statistical

models to identify potentially fraudulent transactions. These systems operate
by applying predetermined rules or algorithms to transaction data, flagging
any deviations from expected behavior as potential fraud cases. While
effective to a certain extent, these systems may struggle to adapt to evolving
fraud tactics and may generate false positives or miss subtle fraudulent
activities.
Additionally, some existing systems utilize machine learning techniques, such
as logistic regression, decision trees, or random forests, to enhance fraud
detection capabilities. These approaches leverage historical transaction data
to train models that can automatically identify patterns indicative of fraudulent
behavior. While more advanced than rule-based systems, machine learning-
based approaches still face challenges in accurately detecting sophisticated
fraud schemes and may require frequent updates to remain effective.

Difference between traditional and Deep learning model

6
4.2 PROPOSED SYSTEM

Our proposed fraud detection system is a big step forward from what's
currently used. Instead of relying on simple rules or old-fashioned methods,
we're using super-smart deep learning technology. This means our system
can learn from lots of transaction data and spot even the sneakiest fraud
patterns in real-time.
To make sure it's really good at catching fraud, we'll train our system on a
bunch of different transactions—some legit, some fraudulent. This helps it
learn what normal transactions look like and what fraud looks like, so it can
tell the difference accurately.
We'll test our system thoroughly to make sure it's super accurate and better
than what's out there already. Once it's ready, it'll be seamlessly integrated
into banks' systems, helping them catch fraud quickly and protect people's
money better than ever before.

7
 CHAPTER 5

FLOW CHART

workflow of fraud detection system

The proposed workflow for the fraud detection system begins with the
collection of transaction data from various sources, including credit card
transactions and banking activities. Once collected, the data undergoes
preprocessing to clean and organize it, ensuring accuracy and consistency.
Subsequently, feature engineering is conducted to select and transform
relevant features that will serve as inputs for the fraud detection model.
Following this, the model is developed using advanced machine learning or
deep learning techniques, trained on the engineered features to accurately
identify fraudulent transactions. Evaluation of the model's performance is then
performed using metrics such as the Area Under the Precision-Recall Curve
(AUPRC) to assess its effectiveness. Upon validation, the model is deployed
into production environments within financial institutions, where it is
continuously monitored to ensure ongoing accuracy and effectiveness in
detecting fraud. Additionally, collaboration with other researchers and
leveraging insights from collaborative research projects allow for continuous
improvement and enhancement of the system's capabilities over time. This
comprehensive workflow aims to provide robust protection against fraudulent
activities within the financial sector.

8
 CHAPTER 6

REQUIREMENT ANALYSIS

PROJECT CATEGORY:

Deep learning (DL)

PROGRAMMING LANGUAGE USED:

PYTHON

LIBRARIEES USED:

• Scikit-learn -for implementing machine learning algorithms

such as MLPClassifier for gender prediction

• NumPy and Pandas- for data manipulation and analysis.

DEVELOPMENT PLATFORM:

Anaconda Prompt (Jupyter Notebook)

DATASET USED:

Credit Card (CSV)

TOOLS USED:

• Editor Used: Anaconda Prompt,

• Operating System: Windows 11
• Processor: Intel core i5 11th
• Gen RAM: 8GB
• Hard Disk: 1TB

9
 CHAPTER 7

IMPLEMENTATION

The implementation phase plays the most important role in the software
development process. It is at this stage that the physical source code of the
system being built is created. Programmers code the IT system on the basis
of the collected requirements and the developed project documentation.

Data Collection and Preprocessing: Gather data relevant to the problem at

hand, such as transaction records, user activities, etc. Preprocess the data
by cleaning, normalizing, and transforming it into a format suitable for deep
learning models. This may involve handling missing values, encoding
categorical variables, and scaling numerical features.

Feature Engineering: Extract meaningful features from the data that can
help the deep learning model distinguish between fraudulent and legitimate
transactions. This might include features such as transaction amount,
location, time of day, frequency of transactions, etc. Feature engineering is
crucial for improving the performance of the model.

Model Selection: Choose an appropriate deep learning architecture for the

task. Common choices include convolutional neural networks (CNNs),
recurrent neural networks (RNNs), and more advanced architectures like
Long Short-Term Memory (LSTM) networks or Gated Recurrent Units
(GRUs). The choice of model depends on the nature of the data and the
complexity of the fraud patterns.

Training the Model: Split the data into training, validation, and testing sets.
Train the deep learning model on the training data using appropriate loss
functions and optimization algorithms (e.g., binary cross-entropy loss and
Adam optimizer). Experiment with different architectures, hyperparameters,
and regularization techniques to improve performance and prevent overfitting.
10
Evaluation: Evaluate the trained model on the validation set to assess its
performance. Common evaluation metrics for fraud detection include
precision, recall, F1-score, and area under the receiver operating
characteristic curve (AUC-ROC). Fine-tune the model based on the validation
results.

Transformation of Data into model

11
7.1 CODING

This Python code builds and trains a neural network model using Keras on
credit card transaction data, aiming to detect fraudulent transactions, followed
by evaluation using metrics such as accuracy, precision, recall, and F1-score.

import numpy as np
import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
from keras.models import Sequential
from keras.layers import Dense, Dropout
from sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score

data = pd.read_csv('creditcard.csv')
print(data)

if 'class' in data.columns:
data.drop(columns=['class'], inplace=True)

X = data.drop(columns=['label']) if 'label' in data.columns else

data.drop(columns=['Class'])
y = data['label'] if 'label' in data.columns else data['Class']

scaler = StandardScaler()
X_scaled = scaler.fit_transform(X)

X_train, X_test, y_train, y_test = train_test_split(X_scaled, y, test_size=0.2,

random_state=42)

model = keras.Sequential([
12
 keras.layers.Dense(64, activation='relu', input_shape=(X_train.shape[1],)),
keras.layers.Dropout(0.2),
keras.layers.Dense(32, activation='relu'),
keras.layers.Dropout(0.2),
keras.layers.Dense(1, activation='sigmoid')])

model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])

history = model.fit(X_train, y_train, epochs=10, batch_size=32, validation_split=0.2)

y_pred_prob = model.predict(X_test)

y_pred = (y_pred_prob > 0.5).astype(int)

accuracy = accuracy_score(y_test, y_pred)

precision = precision_score(y_test, y_pred)
recall = recall_score(y_test, y_pred)
f1 = f1_score(y_test, y_pred)

print("Accuracy:", accuracy)
print("Precision:", precision)
print("Recall:", recall)
print("F1-Score:", f1)

13
 CHAPTER 8

RESULT

result given by the FDS

The result suggests that the fraud detection model achieved high accuracy,
precision, recall, and F1-score:

Accuracy (0.9993): The accuracy indicates that approximately 99.93% of all

transactions were correctly classified by the model. This high accuracy
suggests that the model performs exceptionally well in distinguishing
between fraudulent and non-fraudulent transactions.

Precision (0.8488): Precision measures the proportion of correctly identified

fraudulent transactions among all transactions predicted as fraudulent. In
this case, approximately 84.88% of the transactions predicted as fraudulent
were indeed fraudulent. This indicates that the model has a high level of
reliability when it flags a transaction as potentially fraudulent.

Recall (0.7449): Recall, also known as sensitivity, measures the proportion

of correctly identified fraudulent transactions among all actual fraudulent

14
transactions. A recall of 74.49% suggests that the model successfully
captured 74.49% of all fraudulent transactions present in the dataset.
Although not as high as precision, this level of recall is still considerable and
indicates that the model is effective in identifying fraudulent transactions.

F1-Score (0.7935): The F1-score is the harmonic mean of precision and

recall, providing a balanced measure of the model's performance. With an
F1-score of 0.7935, the model achieves a good balance between precision
and recall. This indicates that the model is effective in both correctly
identifying fraudulent transactions and minimizing false positives.

Overall, the high values of accuracy, precision, recall, and F1-score suggest
that the fraud detection model performs very well on the given dataset,
effectively distinguishing between fraudulent and legitimate transactions.
However, it's essential to consider the specific context of the application and
potential implications of false positives and false negatives for further
evaluation of the model's suitability.

15
 CHAPTER 9

9.1 CONCLUSION

In conclusion, the fraud detection model developed using neural network

techniques exhibits exceptional performance in accurately identifying
fraudulent transactions within credit card data. With an accuracy of 99.93%,
precision of 84.88%, recall of 74.49%, and an F1-score of 79.35%, the model
demonstrates robustness in distinguishing between legitimate and fraudulent
transactions.

This high level of accuracy, coupled with balanced precision and recall,
suggests that the model is reliable and effective in detecting fraudulent
activities. However, further analysis and validation may be necessary to assess
its performance across different datasets and real-world scenarios.

Overall, the results indicate that the developed fraud detection model holds
promise for enhancing security and preventing financial losses in credit card
transactions, showcasing its potential for deployment in real-world financial
systems with appropriate monitoring and continuous improvement practices.

16
9.2 FUTURE WORKS

1. **Enhanced Feature Engineering**: Explore more advanced feature engineering

techniques, such as incorporating temporal patterns, transaction frequency, or
customer behavior analysis, to capture subtle indicators of fraudulent activities
effectively.

2. **Model Optimization**: Investigate advanced neural network architectures,

hyperparameter tuning, and optimization techniques to further improve model
performance and reduce false positives and false negatives.

3. **Imbalanced Data Handling**: Address the issue of imbalanced datasets by

employing techniques such as oversampling, undersampling, or synthetic data
generation to improve the model's ability to detect fraudulent transactions while
mitigating bias.

4. **Anomaly Detection Methods**: Integrate anomaly detection methods alongside

traditional supervised learning approaches to enhance the system's capability to
detect novel and previously unseen fraudulent patterns.

5. **Real-Time Monitoring and Alerts**: Develop real-time monitoring systems that

continuously analyze transaction data streams and trigger alerts or interventions
when suspicious activities are detected, enabling prompt action to prevent fraud.

6. **Interpretability and Explainability**: Enhance the interpretability and explainability

of the model's predictions to provide insights into the factors contributing to fraudulent
transactions, facilitating better decision-making and model refinement.

7. **Adversarial Robustness**: Investigate techniques to enhance the model's

robustness against adversarial attacks and evasion strategies employed by
17
sophisticated fraudsters to deceive the system.

8. **Collaborative Research and Data Sharing**: Foster collaboration with other

researchers and financial institutions to share insights, datasets, and best practices,
facilitating the development of more robust and generalizable fraud detection
systems.

9. **Regulatory Compliance and Ethical Considerations**: Ensure compliance with

regulatory requirements and ethical standards regarding data privacy, security, and
fairness in fraud detection practices.

10. **Continuous Evaluation and Improvement**: Implement mechanisms for

continuous evaluation, monitoring, and refinement of the fraud detection system over
time to adapt to evolving fraud patterns and maintain effectiveness in detecting
fraudulent activities.

18
 CHAPTER 10

REFERENCES:

[1].” Raghavan, P., & El Gayar, N. (2019, December). Fraud detection

using machine learning and deep learning. In 2019 international
conference on computational intelligence and knowledge economy
(ICCIKE) (pp. 334-339). IEEE.

[2].” Pumsirirat, A., & Liu, Y. (2018). Credit card fraud detection using
deep learning based on auto-encoder and restricted boltzmann
machine. International Journal of advanced computer science and
applications, 9(1).

[3].” Pandey, Y. (2017). Credit Card Fraud Detection using Deep

Learning. International Journal of Advanced Research in Computer
Science, 8(5).

[4].” Pillai, T. R., Hashem, I. A. T., Brohi, S. N., Kaur, S., & Marjani, M.
(2018, October). Credit card fraud detection using deep learning
technique. In 2018 Fourth International Conference on Advances in
Computing, Communication & Automation (ICACCA) (pp. 1-6). IEEE.

[5].’ Babu, A. M., & Pratap, A. (2020, December). Credit card fraud
detection using deep learning. In 2020 IEEE Recent Advances in
Intelligent Computational Systems (RAICS) (pp. 32-36). IEEE.

[6].” A.C. Bahnsen, D. Aouada, A. Stojanovic and B. Ottersten,

"Feature engineering strategies for credit card fraud
detection", Expert Syst. Appl., vol. 51, pp. 134-142, 2016.

19
[7].” S. Bhattacharyya, S. Jha, K.K. Tharakunnel and J.C. Westland,
"Combining unsupervised and supervised learning in credit card fraud
detection", Data mining for credit card fraud: A comparative study.
Decis. Support Syst., vol. 50, pp. 602-613, 2011.

[8].” F. Carcillo, Y.L. Borgne, O. Caelen, Y. Kessaci, F. Oblé and G.

Bontempi, "Combining unsupervised and supervised learning in credit
card fraud detection", Inf. Sci., vol. 557, pp. 317-331, 2021.

[9].” Shenvi, P., Samant, N., Kumar, S., & Kulkarni, V. (2019, March).
Credit card fraud detection using deep learning. In 2019 IEEE 5th
International Conference for Convergence in Technology (I2CT) (pp.
1-5). IEEE.

[10].” Beigi, S., & Amin Naseri, M. R. (2020). Credit card fraud
detection using data mining and statistical methods. Journal of AI and
Data Mining, 8(2),149-160.

20