ANOMALY DETECTION FOR FINANCIAL TRANSACTION

SECURITY

A project report submitted in partial fulfilment of the requirements for

the award of the degree of

MASTER OF SCIENCE
IN
INFORMATION TECHNOLOGY

Submitted by

VIGNESH A M
23MIT061

Under the Guidance of

Mrs. SHOBANA., M.Sc., M.Phil., B.Ed.

Assistant Professor,
Department of IT and Cognitive Systems

SRI KRISHNA ARTS AND SCIENCE COLLEGE

An Autonomous Institution,
Accredited by NAAC with ‘A’ Grade
Coimbatore – 641008

OCTOBER 2024

1
 SRI KRISHNA ARTS AND SCIENCE COLLEGE
Accredited by NAAC with ‘A’ Grade, Kuniyamuthur,
Coimbatore – 641008

CERTIFICATE

This is to certify that the project report entitled “ANOMALY DETECTION FOR
FINANCIAL TRANSACTION SECURITY” in partial fulfilment of requirements for the
award of the degree of Master of Science in Information Technology is a record of bonafide work
carried out by VIGNESH A M (23MIT061) and that no part of this has been submitted for
the awardof any other degree or diploma and the work has not been published in popular
journal or magazine.

GUIDE HOD & DEAN

This Project Report is submitted for the viva voce conducted on at Sri
Krishna Arts and Science College.

INTERNAL EXAMINER EXTERNAL EXAMINER

2
 SRI KRISHNA ARTS AND SCIENCE COLLEGE
Accredited by NAAC with ‘A’ Grade, Kuniyamuthur,
Coimbatore – 641008

DECLARATION

I hereby declare that the project report entitled “ANOMALY DETECTION FOR FINANCIAL AND
TRANSACTION SECURITY” submitted in partial fulfilment of the requirements for the award of the
degree of Master of Science in Information Technology is an original work submitted and it has not
been previously formed the basis for the award of any other Degree, Diploma, Associate ship, Fellowship
or similar titles to any other university or body during the periodof my study.

Place: Coimbatore

Date:

Signature of the Candidate

VIGNESH A M

(23MIT061)

3
 ACKNOWLEDGEMENT

I am ineffably indebted to Dr. K. Sundararaman, M.Com., M.Phil., Ph.D.,

CEO, SriKrishna Institutions, Sri Krishna Arts and Science College, Coimbatore.

I convey my profound gratitude to Dr. R. Jagajeevan, MBA., Ph.D.,

Principal, Sri Krishna Arts and Science College for giving me this opportunity to

undergo this Project

It is my prime to solemnly express my sense of gratitude to Dr. K. S. Jeen

Marseline, M.C.A., M.Phil., Ph.D., Head and Dean, Department of IT &

Cognitive Systems, Sri Krishna Arts and Science College.

I would like to extend my thanks and unbound sense for the timely help and

assistancegiven by Mrs. D. Shobana MSC., M.Phil., B.Ed., Assistant Professor,

Department of IT & Cognitive Systems, Sri Krishna Arts and Science College in

completing the Project Report. Her remarkable guidance at every stage of my

work was coupled with suggestion and motivation.

I take this opportunity to thank my parents and friends for their constant

support and encouragement throughout this training.

VIGNESH A M

23MIT061

4
 ABSTRACT

The detection of fraudulent transactions is a critical problem in the financial sec-

tor, requiring accurate and efficient methods to prevent financial losses. This project
presents a deep learning approach to anomaly detection in financial transactions
using a Feed Forward Neural Network (FFNN). The implemented model, a classic
DeepNeural Network (DNN), classifies transactions as fraudulent or sub fraudulent
with impressive accuracy. The model employs a binary classification type with
binary cross-entropy as the loss function and adaptive moment estimation as the
optimizer. The model is trained and tested on a pre-processed and standardized
dataset of credit card transactions. The neural network is built using a Sequential
model, comprising multiple dense layers and a final layer with sigmoid activation
for binary classification. The model is trained using an optimizer and binary cross-
entropy loss function, and evaluated using the accuracy metric. The model achieves
an impressive accuracy of 99.92%, demonstrating its potential for real-world
applications in the financial sector, where accurate and efficient methods are crucial
for preventing financial losses. The high accuracy of the model suggests that it can
be a valuable tool for financial institutions to detect and prevent fraudulent
transactions, ultimately reducing the financial and reputational risks associated with
fraud.

5
 TABLE OF CONTENTS

CHAPTER
CHAPTER TITLE PAGE NO.
NO.

1. INTRODUCTION
1.1 OVERVIEW OF THE PROJECT
1.2 PROBLEM DEFINITION
2. SYSTEM STUDY
2.1 LITERATURE RIVIEW

2.2 PROPOSED SYSTEM

2.3 SYSTEM SPECIFICATION

2.4 SOFTWARE DESCRIPTION
2.5 TECHNOLOGIES ADOPTED
2.6 METHODOLOGIES
3. SYSTEM DESIGN AND DEVELOPMENT
3.1 DATASET STRUCTURE
3.2 USE CASE DIAGRAM
3.3 DATA FLOW DIAGRAM
4. SYSTEM TESTING

5. SYSTEM IMPLEMENATION AND

MAINTENANCE
5.1 RESULT AND DISCUSSION
5.2 CONCULSION

6. BIBLOGRAPHY AND WEB REFERENCES

6.1 SAMPLE CODING

6.2 WEB REFERENCES

6.3 FUTURE ENHANCEMENT

6.4 SAMPLE SCREENSHOTS

6
 CHAPTER 1
INTRODUCTION

In finance, anomaly detection is an important defense against fraudulent

transactions. Every day, financial institutions handle large transactions, and even a
single security breach can result in significant financial losses. This is where deep
neural networks (DNNs) come in as a powerful tool for anomaly detection. DNN
offers un- paralleled accuracy and efficiency in spotting patterns in financial data
and detecting suspicious activity.

Financial practices typically follow certain patterns based on the financial his-
tory of the person or business. By analyzing large amounts of data in advance, DNNs
can recognize these patterns and identify deviations from them. These distractions
can indicate potential fraud, such as unusual purchases, sudden transfers of large
sums of money, or transactions from inaccurate sources. The ability of DNNs to
continuously learn and adapt is an important advantage in anomaly detection. As
fraudsters develop new techniques, DNNs can adapt their models to detect these new
patterns. These changes help ensure that the system remains effective in the face of
evolving threats.

Financial institutions are increasingly using DNN for anomaly detection

due to its effectiveness and versatility. By implementing DNNs, financial
institutions can protect their customers’ funds, reduce fraud losses, and gain
confidence in their financial transactions. It is important to acknowledge that
although DNNs are powerful, they are not a silver bullet. Significant training data
and electronic resources are required to be effective. In addition, it can be difficult
to explain the meaningof DNN’s inconsistency flags. But by working with and
constantly sophisticating human analysts, DNNs are providing a significant boost to
the ongoing fight against financial fraud.

7
1.1 AIM OF THE PROJECT:

The main goal of this project is to harness the power of deep neural networks
(DNNs) to develop a robust anomaly detection system for financial transactions This
system will be built to analyze more financial data and identify deviant patterns from
the norm. These deviations can indicate fraud, such as money laundering or
unauthorized accounts. By implementing this DNN-based system, we aim to
significantly enhance the security and efficiency of financial institutions. Early
identification of discrepancies can lead to early intervention, reducing financial loss
and protecting consumer assets. Furthermore, the system’s ability to constantly learn
and adapt will make it more effective against evolving fraudulent techniques.

1.2 PROJECT DOMAIN:

This work delves into the use of deep neural networks (DNNs) for anomaly
detection in financial transactions. DNNs, with their unique pattern recognition
capabilities, are ideally suited for analyzing the vast amount of data generated by
financial institutions on a daily basis Through the specific process of identifying
relevant transactions based on historical data on it, DNNs can appropriately flag
obstacles thatmay indicate fraudulent activity. The goal of this project is to develop
a DNN-based system that can identify suspicious transactions in real-time, which
can intervene faster and protect financial institutions and their customers from
fraud. The focus of this project will be to build and train a DNN model specially
designed for anomaly detection in financial transactions. The project will explore
different DNN algorithms and training methods to improve the accuracy and
performance of the model. In addition, the project will address the challenges of
interpretation and bias arising with DNNs. By using optimal translation methods
and incorporating fairness considerations throughout the development process, the
project seeks to develop robust and reliable DNN algorithms for real-world
economic applications.

8
1.3 SCOPE OF THE PROJECT:

The objective of this project is to develop a deep neural network (DNN) model
for anomaly detection in financial transactions. The model will be trained on
historical data on financial transactions to identify specific examples of appropriate
use. By analyzing incoming actions in real time, DNN detects deviations from these
norms that could indicate potential fraud.

The scope of the project covers the entire development life cycle of DNN.
This includes data acquisition and pre-processing, design and training of the DNN
system, and integration of the system into a real-time anomaly detection system The
project will also develop performance metrics to assess the use of the DNN system
effectively in detecting fraudulent transactions.

PURPOSE

The purpose of this project on anomaly detection for financial transaction

security is to develop a sophisticated model that identifies unusual patterns in
transaction data, thereby enhancing fraud prevention measures. In today’s digital
financial landscape, the volume of transactions is vast, making it challenging to monitor
and detect fraudulent activities manually. By utilizing Feedforward Neural Networks
(FFNN) and Deep Neural Networks (DNN), this project aims to automate the detection
process, enabling real-time analysis of transactions and improving response times to
potential fraud. The model will be trained on historical transaction data, learning to
differentiate between legitimate transactions and those that exhibit anomalous behavior.
This not only safeguards financial institutions and their customers but also fosters trust
in digital transactions. Ultimately, the project seeks to provide an efficient, scalable
solution that enhances security measures, reduces financial losses, and contributes to
the overall integrity of the financial system.

9
 CHAPTER 2
SYSTEM STUDY

2.1 LITERATURE REVIEW

EXISTING SYSTEM:

Existing systems for anomaly detection in financial transactions primarily rely

on rule-based approaches and traditional machine learning techniques. Many
institutions use fixed rules to flag suspicious transactions based on predefined
thresholds, such as unusually large amounts or transactions occurring in rapid
succession. While effective for some straightforward cases, these systems often struggle
to adapt to evolving fraud patterns and can result in a high rate of false positives, leading
to customer dissatisfaction and operational inefficiencies.

These models require extensive feature engineering and may not capture
complex relationships within the data. However, they typically lack the depth needed to
recognize sophisticated fraud patterns that deep learning methods can address.

Some advanced systems incorporate ensemble methods, combining multiple

algorithms to improve accuracy. Nevertheless, they still face limitations in real-time
processing capabilities and scalability. Overall, while existing systems provide a
foundational approach to fraud detection, they often fall short in adaptability and
efficiency compared to deep learning techniques, highlighting the need for a more
robust, real-time solution that leverages modern machine learning frameworks like
FFNN and DNN.

2.2 PROPOSED SYSTEM:

The proposed system for the anomaly detection project in financial transaction
security aims to create a robust, scalable solution capable of identifying fraudulent
activities in real-time. The system will utilize Feedforward Neural Networks (FFNN)
and Deep Neural Networks (DNN) to analyse transaction data effectively. Initially, the
system will gather and preprocess large datasets containing historical transaction
records, ensuring data quality through techniques like normalization and feature

10
extraction. Key features, such as transaction amount, time of transaction, user location,
and historical spending patterns, will be incorporated to improve model accuracy.
Once the data is prepared, the model will be trained using labeled datasets, allowing it
to learn the characteristics of normal and anomalous transactions. A validation phase
will ensure that the model generalizes well to unseen data. The system will be designed
to operate in a real-time environment, enabling immediate detection and flagging of
suspicious transactions for further investigation. Additionally, a user-friendly dashboard
will provide visual insights into transaction patterns and model performance metrics.
By implementing this system, financial institutions can enhance their fraud detection
capabilities, minimize losses, and bolster customer trust in digital transaction security.
This proactive approach aims to significantly improve the overall security landscape of
financial transactions.

2.3 FEASIBILITY STUDY:

The feasibility study for the anomaly detection project in financial transaction
security evaluates its technical, operational, and economic viability. Technically, using
Feedforward Neural Networks (FFNN) and Deep Neural Networks (DNN) is viable, as
these models can effectively analyze large datasets and complex transaction patterns.
Operationally, integration with existing financial systems is achievable, requiring
collaboration with IT teams for real-time data processing. Economically, although initial
costs for development may be high, the long-term benefits—such as reduced fraud
losses and increased customer trust—justify the investment. Overall, the proposed
system promises significant enhancements in detecting fraudulent transactions.

2.3.1 ECONOMIC FEASIBILITY:

The proposed project is economically feasible. The cost of hosting the

projecton a cloud platform is relatively low, and the cost of development can be
managedby a small team of developers. Additionally, the potential benefits of the
system, such as reducing losses due to Anomaly, outweigh the costs of development
and maintenance.

11
2.3.2 TECHNICAL FEASIBILITY:

The technical feasibility of the anomaly detection project hinges on the

utilization of Feedforward Neural Networks (FFNN) and Deep Neural Networks
(DNN), which are well-suited for processing large datasets and capturing complex
patterns in transaction behavior. Existing machine learning frameworks, such as
TensorFlow and PyTorch, provide robust tools for model development and training.
The system requires reliable data sources for historical transaction data and efficient
preprocessing techniques to ensure data quality. With adequate computational
resources, including GPUs for model training, the implementation of this system is
technically achievable, promising improved detection capabilities in financial
transactions.

2.3.3 SOCIAL FEASIBILITY

The social feasibility of the anomaly detection project is strong, as it addresses

the growing concern over financial fraud, enhancing consumer trust in digital
transactions. By improving security measures, the system can help protect users’
sensitive information and financial assets, contributing to overall public confidence in
online banking. Additionally, increased awareness and education about fraud prevention
can empower users, fostering a proactive community approach to financial security.

2.4 SYSTEM SPECIFICATION

SOFTWARE SPECIFICATION

Operating System : Windows 10/11, macOS, or a Linux distribution.

Programming Language : Python

12
HARDWARE SPECIFICATION

Processor:
• Intel i7 or AMD Ryzen 7 (or higher) for better performance with deep learning
tasks.
• At least 8 cores recommended.
RAM:
• Minimum 16 GB, ideally 32 GB for handling large datasets and training models
efficiently.
Storage:
• SSD (Solid State Drive) with at least 512 GB for faster data access and
processing.
• Consider an additional HDD for larger datasets.
GPU:
• NVIDIA RTX 2060 or higher (e.g., RTX 3060, 3070, etc.) for accelerated
training of neural networks. CUDA support is essential for GPU acceleration.

2.4.1 SOFTWARE DESCRIPTION

NUMPY:

NumPy, short for Numerical Python, is a powerful library that serves as the
backbone for numerical computing in Python. It provides a high-performance
multidimensional array object, along with tools for working with these arrays. NumPy’s
core feature is the ndarray, which enables efficient storage and manipulation of large
datasets. It supports a wide range of mathematical functions, making it ideal for
operations like linear algebra, Fourier transforms, and random number generation.
NumPy’s broadcasting capabilities allow for arithmetic operations between arrays of
different shapes, enhancing flexibility. In the context of data science and machine
learning, NumPy is often used for preprocessing data, performing calculations, and
serving as a foundation for other libraries, such as Pandas and TensorFlow. Its speed
and efficiency make it an essential tool for anyone working with numerical data,
facilitating complex computations with ease and precision.

13
PANDAS:

Pandas is a powerful data manipulation and analysis library for Python, designed
to make working with structured data intuitive and efficient. At its core, Pandas
introduces two primary data structures: Series (one-dimensional) and Data Frame (two-
dimensional), which allow users to easily handle and analyze tabular data. With built-
in functions for data cleaning, transformation, and aggregation, Pandas simplifies tasks
such as handling missing values, filtering data, and performing group operations. Its
ability to read and write data from various formats, including CSV, Excel, and SQL
databases, makes it highly versatile for data ingestion. In the realm of data analysis and
machine learning, Pandas is frequently used for preprocessing datasets before modeling.
Its seamless integration with NumPy enhances numerical computations, while its rich
visualization capabilities, in conjunction with libraries like Matplotlib and Seaborn,
provide valuable insights into data trends and patterns, making it indispensable for data
scientists and analysts alike.

DNN:

Deep Neural Networks (DNNs) are a class of artificial neural networks

characterized by multiple layers of interconnected nodes, or neurons. Each layer
transforms the input data through weighted connections and activation functions,
enabling DNNs to learn complex patterns and representations. DNNs excel in tasks such
as image recognition, natural language processing, and anomaly detection due to their
capacity to model intricate relationships within large datasets. By stacking layers, DNNs
can capture hierarchical features, allowing them to perform well on diverse tasks.
Training a DNN involves optimizing weights using algorithms like backpropagation,
requiring substantial computational resources, especially for deep architectures.

FFNN:

Feedforward Neural Networks (FFNNs) are a type of artificial neural network

where connections between nodes do not form cycles. In an FFNN, data flows in one
direction—from input layers, through hidden layers, to the output layer. Each neuron
processes inputs using weighted sums and an activation function, allowing the network
to learn complex patterns in data. FFNNs are particularly effective for tasks such as

14
classification and regression. They are simpler than deep neural networks, making them
easier to implement and train. However, they may struggle with highly complex tasks,
where deeper architectures often yield better performance.

2.6 TECHNOLOGIES ADOPTED

PYTHON:

Python is a versatile, high-level programming language known for its readability

and ease of use. Developed in the late 1980s, it emphasizes simplicity and code clarity,
making it an ideal choice for both beginners and experienced developers. Python
supports multiple programming paradigms, including procedural, object-oriented, and
functional programming, which enhances its flexibility in various applications. Its
extensive standard library and rich ecosystem of third-party packages, such as NumPy,
Pandas, and TensorFlow, make it particularly popular in data science, machine learning,
web development, and automation. Python's active community contributes to its
continuous evolution, providing extensive documentation, resources, and frameworks
that facilitate development. Additionally, Python's compatibility with different
platforms and integration capabilities with other languages allow it to be used in diverse
environments. This combination of features has solidified Python's position as one of
the leading programming languages in the tech industry, driving innovation and
collaboration across various fields.

MACHINE LEARNING:

Machine learning is a subset of artificial intelligence that focuses on developing

algorithms that enable computers to learn from and make predictions based on data.
Instead of being explicitly programmed, machines improve their performance through
experience, identifying patterns and insights within large datasets. Machine learning
encompasses various techniques, including supervised learning, unsupervised learning,
and reinforcement learning, each suited for different types of tasks. Applications span
numerous fields, such as finance, healthcare, and natural language processing, where
machine learning enhances decision-making, automates processes, and drives
innovations.

15
2.7 METHODOLOGIES

The methodology for your anomaly detection project in financial transaction

security begins with data collection, where you gather a dataset comprising both normal
and anomalous transactions. Next, you preprocess the data by addressing missing
values, normalizing features, and encoding categorical variables to ensure consistency.
Following this, feature selection is performed to identify key attributes that influence
transaction behavior, utilizing techniques like correlation analysis. You will then
develop models using Feedforward Neural Networks (FFNN) and Deep Neural
Networks (DNN) for classification. The dataset is split into training and testing sets,
employing cross-validation to assess model performance. Finally, the trained models
are applied to detect and flag unusual transactions, thereby enhancing security measures.

Implementation of Feedforward Neural Networks (FFNN)

Feedforward Neural Networks (FFNN) are structured layers of interconnected neurons

where data flows in one direction, facilitating efficient processing and learning of complex
patterns in financial transaction datasets for anomaly detection.

Implementation of Deep Neural Networks (DNN)

Deep Neural Networks (DNN) consist of multiple hidden layers that capture intricate
features and relationships in data, enhancing the model's ability to detect subtle anomalies in
financial transactions through hierarchical representation learning.

16
 CHAPTER 3
SYSTEM DESIGN AND DEVELOPMENT

3.1 DATASET STRUCTURE:

The dataset for your anomaly detection project in financial transaction security should
consist of various features representing transaction attributes. Key fields may include
transaction ID, timestamp, user ID, transaction amount, transaction type (e.g., withdrawal,
deposit), merchant details, location, and device used. Additionally, it is essential to include
labels indicating whether a transaction is normal or anomalous. The dataset should contain a
diverse range of transactions to capture different spending behaviors and anomalies. A
balanced representation of both legitimate and fraudulent transactions will enhance the model's
ability to learn distinguishing features effectively, improving anomaly detection accuracy.

3.2 USE CASE DIAGRAM:

A use case diagram is a visual representation that illustrates the interactions between
various actors and the system in a project. In the context of your anomaly detection project for

17
financial transactions, the diagram serves to identify the key functionalities of the system.

3.3 DATA FLOW DIAGRAM:

A Data Flow Diagram (DFD) visually represents the flow of data within a system. In
your anomaly detection project, it shows how transaction data is input by customers, processed
by the fraud detection system, and outputs flagged transactions to the bank admin for review,
facilitating efficient data handling and security.

3.4 ACTIVITY DIAGRAM:

An Activity Diagram visually represents the dynamic aspects of a system by outlining

workflows and processes involved in anomaly detection for financial transactions. In your
project, the diagram illustrates a sequence of actions that begin when a bank customer initiates
a transaction. The fraud detection system then monitors this transaction in real-time for any
anomalies. If an irregularity is detected, the system flags the transaction as suspicious. This
triggers a review process by the bank admin, who examines the flagged transaction details.
Finally, the admin can generate reports summarizing findings, ensuring clarity in operations
and enhancing security measures against fraud.

18
3.5 CLASS DIAGRAM

Shows that the Anomaly detection system includes several classesthat represent the
different components of the system, including the Transaction, User, Feature Engineering,
Machine Learning, Fraud Detector, Fraud Investigation, and Notification classes.

19
 CHAPTER 4

SYSTEM TESTING

In my anomaly detection project for financial transactions, system testing plays a

crucial role in verifying the effectiveness and reliability of the developed solution. This phase
involves comprehensive evaluations to ensure the system meets specified requirements for
detecting fraudulent activities. Testing types, such as functional testing, assess the system's
ability to accurately identify anomalies in various transaction scenarios. Performance testing
evaluates the system's responsiveness under high transaction volumes, while security testing
ensures sensitive data is protected

TESTING TYPES:

UNIT TESTING:

Unit testing focuses on verifying the functionality of individual components within the
anomaly detection system, such as algorithms for detecting fraudulent transactions and data
processing functions. By testing each unit in isolation, developers can identify and fix bugs
early, improving code quality and facilitating smoother integration into the overall system
architecture

20
INTEGRATION TESTING:

Integration testing evaluates the interaction between different components of the

anomaly detection system, such as the transaction input module and the fraud detection
algorithms.This testing ensures that these modules work together as intended,

from your module i m p o r t Anomaly Detector

s e l f . d e t e c t o r = Anomaly Detector ( )

identifying issues in data flow and

communication, thereby enhancing the system's overall functionality and reliability.

21
import u n i t t e s t
from your module i m p o r t Anomaly
Detector from your m o d ul e i m p o r t D a t a
P r e p r o c e s s o rfrom your m o d ul e i m p o r t
DataGenerator

c l a s s T e s t I n t e g r a t i o n ( u n i t t e s t . T e s t C a s e ) :d e f set Up ( s e l f ) :
s e l f . d e t e c t o r = Anomaly Detector ( )
se l f . preproce ssor = DataPrepro
c e s s o r ( )s e l f . g e n e r a t o r = D a t a G e n e
rator ( )

def test integration

( s e l f ) : # Load t h e d
ata
data = s e l f . generator . ge ne rate data ( )

# Preprocess the data

preprocesse d data = se l f . preprocessor . preprocess ( data )

# T r a i n t h e model
sel f . detector . t rai n ( preprocessed data )

# Make p r e d i c t i o n s
predictions = sel f . detector . predict ( preprocessed data )

# E v a l u a t e t h e model
evaluation = sel f . detector . evaluate ( preprocessed data )

# Check t h e results
se l f . assert Greater ( evaluat
i o n , 0 )s e l f . a s s e r t L e s s ( e v a
l u a t i o n , 1)

# Check t h e p r e d i c t i o n s
se l f . assertEqual ( len ( p redi cti ons ) , len ( pre processed
d a t a ) )s e l f . a s s e r t G r e a t e r E q u a l ( max ( p r e d i c t i o n s ) , 0 )
s e l f . a s s e r t L e s s E q u a l ( min ( p r e d i c t i o n s ) , 1 )

if name == ’m a i n ’ :un i t t e s t . main )

SYSTEM TESTING:

System testing for the anomaly detection project involves evaluating the entire
application to ensure it meets specified requirements. This includes functional testing to verify
accurate anomaly detection, performance testing for responsiveness under load, security testing

22
to identify vulnerabilities, and user acceptance testing to confirm usability, ultimately ensuring
a robust and reliable system.

import u n i t t e s t
from your module i m p o r t Anomaly Detector
from your m o d ul e i m p o r t D a t a P r e p r o c e s s o r
from your m o d ul e i m p o r t D a t a G e n e r a t o r

c l a s s Te st Sy st e m ( u n i t t e s t . T e s t C a s e ) :d e f set Up ( s e l f ) :
s e l f . d e t e c t o r = Anomaly Detector ( )
s e l f . p r e p r o c e s s o r = D a t a P r e p r o c e s s o r ( )s e l f . g e n e r a t o r = D a t a G e n e r a t o r ( )
d e f t e s t s y s t e m ( s e l f ) :# Load t h e d a t a
data = se l f . gene rator . ge ne rate da ta ()
# Preprocess the data
preprocesse d data = se l f . preprocessor . preprocess ( data )
# T r a i n t h e model
sel f . detector . t rai n ( preprocessed data )
# Make p r e d i c t i o n s
predictions = sel f . detector . predict ( preprocessed data )

# E v a l u a t e t h e model
evaluation = se lf . detector . evaluate ( preprocessed data )

# Check t h e r e s u l t s
s e l f . a s s e r t G r e a t e r ( e v a l u at i o n , 0 )
s e l f . a s s e r t L e s s ( e v a l u a t i o n , 1 )a
# Check t h e p r e d i c t i o n s

se l f . assertEqual ( len ( p redi cti ons ) ,

len ( preprocessed data ) )

s e l f . a s s e r t G r e a t e r E q u a l ( max ( p r e d i c t i o n s ) , 0 )
s e l f . a s s e r t L e s s E q u a l ( min ( p r e d i c t i o n s ) , 1 )
# T e s t t h e anomal y d e t e c t i o n
anomalies = s e l f . d e t e c t or . de t e ct a noma l i e s ( pre pr oc e sse d da t a )
s e l f . a s se rt Equal ( len ( anomalies ) , len ( p re proc e sse d da ta ) )
s e l f . a s s e r t G r e a t e r E q u a l ( max ( a n o m a l i e s ) , 0 )
s e l f . a s s e r t L e s s E q u a l ( min ( a n o m a l i e s ) , 1 )
if name == m a i n ’ : u n i t t e s t . main ( )

23
4.1 INPUT MODEL:

The input model for your anomaly detection system is designed to capture essential
features from financial transaction data, enabling accurate analysis. Key inputs include
transaction ID, timestamp, user ID, transaction amount, transaction type (e.g., withdrawal,
deposit), merchant details, geographical location, and device information used for the
transaction. These features collectively provide a comprehensive view of each transaction,
allowing the system to identify patterns and detect anomalies effectively. To optimize model
performance, the input data must undergo preprocessing.

4.3 OUTPUT MODEL:

The anomaly detection algorithm in your financial transaction system employs machine
learning techniques to identify irregular patterns. Initially, the algorithm processes input
features such as transaction amount, timestamp, user ID, and transaction type. Common
algorithms include Isolation Forest, which isolates anomalies by randomly partitioning data,
and One-Class SVM, which learns the boundaries of normal transactions. K-Means Clustering
can also be used to group similar transactions, identifying outliers as anomalies. After training,
the model assigns risk scores to transactions, flagging those exceeding a predefined threshold.
This systematic approach enhances fraud detection, providing accurate alerts for suspicious
activities and improving financial security.

24
 CHAPTER 5

SYSTEM IMPLEMENTATION AND MAINTAINENCE

5.1 ENVIRONMENT SETUP:

To set up the environment for your anomaly detection project, install Python as the
primary programming language, along with essential libraries like Pandas for data
manipulation, Scikit-learn for machine learning, and Matplotlib/Seaborn for data visualization.
Additionally, configure a database (e.g., MySQL or MongoDB) to store transaction data, and
consider using Jupyter Notebook for interactive development and testing.

DATA COLLECTION AND PREPROCESSING:

• Data Acquisition: Gather historical transaction data from relevant sources, such as
bank databases or financial APIs, ensuring the dataset includes both legitimate and
fraudulent transactions for comprehensive analysis.
• Data Cleaning: Remove duplicates, handle missing values, and correct inconsistencies
in the dataset to ensure high-quality data for analysis. This may involve imputing
missing values or removing problematic entries.
• Feature Selection: Identify and select relevant features, such as transaction ID,
timestamp, user ID, transaction amount, transaction type, and merchant details, that
contribute to detecting anomalies.
• Normalization and Encoding: Normalize numerical features to a consistent scale
(e.g., using Min-Max scaling) and encode categorical variables (e.g., using one-hot
encoding) to prepare the data for machine learning algorithms.
• Data Splitting: Divide the pre-processed dataset into training, validation, and testing
subsets to facilitate model training, tuning, and evaluation, ensuring that the model
generalizes well to unseen data.

25
5.2 DATA COLLECTION AND PREPROCESSING:

The dataset as shown in contains transactions made by credit cards in September 2013
by European cardholders. This dataset presents transactions that occurred in two days, where
we have 492 frauds out of 284,807 transactions. The dataset is highly unbalanced, the positive
class (frauds) account for 0.172% of all transactions.

It contains only numerical input variables which are the result of a PCA trans- formation.
Unfortunately, due to confidentiality issues, we cannot provide the original features and more
background information about the data. Features V1, V2, . . . V28 are the principal components
obtained with PCA, the only features which have not been transformed with PCA are ’Time’
and ’Amount’. Feature ’Time’ contains the seconds elapsed between each transaction and the
first transaction in the dataset. The feature ’Amount’ is the transaction Amount, this feature
can be used for example dependent cost-sensitive learning. Feature ’Class’ is the response
variable and it takes value 1 in case of fraud and 0 otherwise.

26
5.3 MODEL EVALUATION AND DEPLOYMENT:

In this module, we will evaluate the performance of the trained FFNN model
using metrics such as precision, recall, and F1-score. We will also use techniques
such as confusion matrices and ROC curves to visualize the performance of the
model. If the model performs well, we will deploy it to detect anomalies in real-time
financial transactions. We will also monitor the performance of the model and
retrain it peri- odically to adapt to changes in the data distribution. Additionally, we
will integrate the model with other systems, such as fraud detection systems, to
improve the overall accuracy of the system.

5.4 STEPS TO EXECUTE/RUN/IMPLEMENT THE PROJECT:

LOAD THE DATASET:

Start by loading the credit card transaction data from the ‘creditcard.csv
‘dataset in Google Colab. This dataset will be split into training and testing sets.
Ensure to remove any missing data during this process.

EXPLORE THE DATASET:

Explore the dataset to understand the structure and distribution of the data. Pay attention
to relevant features such as transaction amount, location, and time. Remove any irrelevant or
redundant features to streamline the data.

27
PREPROCESS THE DATASET:

Preprocess the data to prepare it for training. This may include steps such as scaling,
normalization, and handling missing data. The data should be labeled appropriately, where
fraudulent transactions are labeled as 1 and non-fraudulent trans- actions as 0.

TRAIN THE MODEL:

Train a Deep Neural Network model on the training set. This involves computing the
sigmoid function on the linear combination of features and weights, computing the loss
function using the predicted probabilities and the true labels, computing the gradient of the loss
function with respect to the weights, and updating the weights using a learning rate and the
gradient.

TEST THE MODEL:

Apply the trained Deep Neural Network model to the testing data and evaluate its
performance using metrics such as accuracy, precision, recall, and F1 score.

EXISTING SYSTEM AND PROPOSED SYSTEM

EXISTING SYSTEM:

The existing anomaly detection system relies on a combination of machine learning

algorithms, including logistic regression, to identify potential fraudulent transactions. Unlike
traditional rule-based systems that rely on predefined rules and thresholds, this system uses
machine learning to analyze complex patterns and relationships in the data. For instance, if a
transaction exhibits unusual characteristics, such as being significantly larger than the average
transaction for that customer, the system may flag it as potentially fraudulent. Similarly, if a
transaction occurs in a country outside the customer’s typical travel pattern, the system may
raise an alert. These alerts are then reviewed and investigated by human analysts, who use their
expertise to determine whether the transaction is indeed fraudulent or not.

28
 One limitation of the existing system is its reliance on predefined rules and thresholds,
which can limit its effectiveness in detecting new or evolving fraud pat- terns. Additionally,
the existing system may generate false positives or false negatives, which can lead to
unnecessary investigations or missed fraudulent transactions

COMPARISON GRAPH:

COMPARISON TABLE:

Algorithm Accuracy (Training Set) Accuracy (Testing Set)

Deep Neural Network 0.99 0.99
Support Vector Machine 0.92 0.89
Random Forest 0.88 0.84
Logistic Regression 0.85 0.81
K-Nearest Neighbors 0.80 0.78

PROPOSED SYSTEM:

The proposed anomaly detection system utilizes a feedforward neural network

29
(FFNN) to analyze vast amounts of credit card transaction data and uncover complex
patterns indicative of fraudulent activity. The FFNN is trained on a combination of
advanced machine learning techniques, including logistic regression, decision trees, and
random forests, allowing the system to continuously learn and refine its accuracy over
time. By incorporating new data and user feedback into its algorithms, the system can
adapt to evolving fraudulent tactics and improve its detection capabilities. One of the
system’s key advantages is its ability to detect fraud in real-time, enabling swift
intervention to prevent losses and minimize the impact of fraudulent transactions.
Additionally, the system automates a significant portion of the fraud detection process,
reducing the need for manual reviews and investigations, and subsequently decreasing
associated costs.

Furthermore, the FFNN’s ability to learn and generalize from large datasets
enables the system to identify subtle patterns and anomalies that may not be detectable
by traditional rule-based systems. However, a potential challenge lies inthe system’s
reliance on substantial volumes of high-quality data to effectively train its FFNN.
Furthermore, like any detection system, there is a risk of false positivesor false
negatives, potentially leading to unnecessary investigations or overlooked fraudulent
activities.

30
 CHAPTER 6

CONCLUSION:

In conclusion, the anomaly detection project for financial transactions addresses a

critical need for enhanced security in banking operations. By leveraging machine learning
algorithms, such as Isolation Forest and Neural Networks, the system effectively identifies
fraudulent activities by analysing patterns in transaction data. The comprehensive approach to
data collection and preprocessing ensures that the model is trained on high-quality, relevant
information, enhancing its accuracy and reliability. Rigorous testing phases, including unit,
integration, performance, and security testing, validate the system’s functionality and
robustness, ensuring it performs well under various conditions and remains resilient against
potential threats. The user-friendly interface allows bank administrators to easily review
flagged transactions and generate insightful reports, facilitating timely interventions and
informed decision-making. Ultimately, this project not only improves the detection of
anomalies but also fosters trust and confidence among users by safeguarding their financial
transactions. As the system evolves, ongoing monitoring and updates will be essential to adapt
to emerging threats and changing transaction patterns. This proactive stance will help maintain
the system's effectiveness in combating fraud, ensuring it remains a vital tool for enhancing
financial security in an increasingly complex digital landscape.

Looking ahead, further enhancements could involve integrating advanced techniques

such as deep learning and real-time analytics to improve detection capabilities. Additionally,
expanding the dataset to include diverse transaction scenarios will bolster the model's training,
allowing it to adapt to new fraud patterns. Collaborating with cybersecurity experts will also
provide insights into evolving threats, ensuring that the system remains at the forefront of
financial security measures. Ultimately, this project sets the foundation for ongoing innovation
in protecting financial transactions.

31
FUTURE ENCHANCEMENTS:

Future enhancements for the anomaly detection system in financial transactions can
significantly improve its effectiveness and adaptability to emerging threats. One potential
enhancement is the integration of advanced machine learning techniques, such as deep learning
models like Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs).
These models can capture more complex patterns in transaction data, improving the system's
ability to identify subtle anomalies that traditional algorithms may overlook. Additionally,
incorporating ensemble methods, which combine multiple models to improve prediction
accuracy, could further enhance detection capabilities.

Another important area for enhancement is the expansion of the dataset used for training
the model. By incorporating a more diverse range of transaction scenarios—including different
geographical locations, user behaviours, and transaction types—the model can become more
robust and better equipped to recognize fraudulent activities across various contexts. Regularly
updating the dataset with recent transactions will also help the model adapt to evolving fraud
\tactics.

Furthermore, integrating real-time analytics can provide immediate insights into

transaction patterns, allowing for faster detection and response to potential anomalies.
Implementing streaming data processing frameworks will enable the system to analyze
transactions as they occur, significantly reducing response times and improving overall
security.

Collaboration with cybersecurity experts can also provide valuable insights into
emerging threats and vulnerabilities. By staying informed about the latest fraud trends and
attack vectors, the system can be proactively updated to mitigate risks effectively. Additionally,
incorporating user feedback loops will help refine the model based on real-world experiences,
making the system more user-centric.

Lastly, enhancing user interfaces for bank administrators will improve usability.
Providing intuitive dashboards, customizable alerts, and detailed visualizations of transaction
patterns will empower users to make informed decisions quickly.

32
 CHAPTER 7

BIBLIOGRAPHY:

• Ahmed, M., Mahmood, A. N., & Hu, J. (2016). A survey of network anomaly detection
techniques. Journal of Network and Computer Applications, 60, 1-22.
https://doi.org/10.1016/j.jnca.2015.09.015
• Chandola, V., Banerjee, A., & Kumar, V. (2009). Anomaly detection: A survey. ACM
Computing Surveys (CSUR), 41(3), 1-58. https://doi.org/10.1145/1541880.1541882
• Iglewicz, B., & Hoaglin, D. C. (1993). How to Detect and Handle Outliers. New York:
Sage Publications.
• Kiran, R. A., & Gohil, H. (2017). A comparative study of machine learning algorithms
for anomaly detection in financial transactions. International Journal of Computer
Applications, 168(2), 1-5. https://doi.org/10.5120/ijca2017915260
• Xia, Y., Wang, L., & Wu, Y. (2015). Financial transaction anomaly detection based on
machine learning algorithms. Journal of Financial Crime, 22(3), 350-363.
https://doi.org/10.1108/JFC-04-2014-0022
• Wu, J., & Zhang, J. (2018). Anomaly detection in financial transactions: A deep
learning approach. Expert Systems with Applications, 95, 174-183.
https://doi.org/10.1016/j.eswa.2017.11.052
• Zhou, Z., & Jiang, Y. (2019). A survey of anomaly detection techniques in financial
transactions. Journal of Financial Crime, 26(2), 336-352. https://doi.org/10.1108/JFC-
10-2018-0106
• Bhatia, A., & Singh, M. (2020). Machine learning for fraud detection: A review.
Journal of King Saud University - Computer and Information Sciences.
https://doi.org/10.1016/j.jksuci.2020.03.005

33
 CHAPTER 8

8.1 SAMPLE CODING:

from g o o g l e . c o l a b
i m p o r t d r i v ed r i v e .
mount ( ’ / c o n t e n t / d r
ive ’ )

import
pandas as
pd i m p o r t
numpy as np
import mat pl otlib .
p y p l o t as p l t i m p o r t
s e a b o r n as s n s
from s k l e a r n . p r e p r o c e s s i n g i m p o r t S t a n d a r d S c a l e r

from s k l e a r n . m o d e l s e l e c t i o n i m p o r t
t r a i n t e s t s p l i ti m p o r t k e r a s
from k e r a s . models i m p o
r t S e q u e n t i a l from k e r a s
. l a y e r s i m p o r t Dense
from k e r a s . l a y e r s i m p o r t Dropout

d r i v e . mount ( ’ / c o n t e n t / d r i v e ’ )

i m p o r t os

os . l i s t d i r ( ’ / c o n t e n t / d r i v e / MyDrive / Colab Notebooks ’ )

df = pd . r e a d c s v ( ’ / c o n t e n t / d r i v e / MyDrive / Colab Notebooks / c r e d i t c a r d . csv ’ )

df . head ( 1 )

34
 df [ ’ C l a s s ’ ] . u n i q u e ( ) # 0 = no f r a u d , 1 = f r a u d u l e n t

p r i n t ( df . shape )

p r i n t ( df . i n f o ( ) )

p r i n t ( df . d e s c r i b e ( ) )

s n s . c o u n t p l o t ( x= ’ C l a
s s ’ , d a t a = df ) p l t . show
()

X = df . i l o c [ : , : − 1 ] . v a l u e s

y = df . i l o c [ : , − 1 ] . v a l u e s

p r i n t (X. shape )

p r i n t ( y . shape )

X t r a i n , X t e s t , Y t r a i n , Y t e s t = t r a i n t e s t s p l i t ( X, y , t e s t s i z e = 0 . 1 , r a n
d o m s t at e =1)

sc = S t a n d a r d S c a l e r ( )

X t r a i n = sc . f i t t r a n s f o r m ( X t r a i n )

X t e s t = sc . t r a n s f o r m ( X t e s t )

p r i n t ( X t r a i n . shape

p r i n t ( X t e s t . shape )

clf = Sequential ([

Dense ( u n i t s = 16 , k e r n e l i n i t i a l i z e r = ’ uniform ’ , i n p u t d i m = 30 , a c t i v a t i o n = ’ r e l u ’

35
 ),

Dense ( u n i t s = 18 , k e r n e l i n i t i a l i z e r = ’ uniform ’ , a c t i v a t i o n = ’ r e l u ’ ) ,

Dropout ( 0 . 2 5 ) ,

Dense ( 2 0 , k e r n e l i n i t i a l i z e r = ’ uniform ’ , a c t i v a t i o n = ’ r e l u ’ ) ,

Dense ( 2 4 , k e r n e l i n i t i a l i z e r = ’ uniform ’ , a c t i v a t i o n = ’ r e l u ’ ) ,

Dense ( 1 , k e r n e l i n i t i a l i z e r = ’ uniform ’ , a c t i v a t i o n = ’ s i gmoid ’ )] )

c l f . summary ( )

c l f . compile ( o p t i m i z e r = ’ adam ’ , l o s s = ’ b i n a r y c r o s s e n t r o p y ’ , m e t r i c s =[ ’ a c c u r a c y
’ ])

c l f . f i t ( X t r a i n , Y t r a i n , b a t c h s i z e = 15 , epochs = 2 )

s c o r e = c l f . e v a l u a t e ( X t e s t , Y t e s t , b a t c h s i z e = 128 )

p r i n t ( ’ \nAnd t h e Score i s ’ , s c o r e [ 1 ] * 100 , ’%’

# Predicting the te st set results

y pred = clf . predict ( X test )

y pred = ( y pred > 0.5)

from s k l e a r n . m e t r i c s i m p o r t c o n f u s i o n m a t r i x

from s k l e a r n . m e t r i c s i m p o r t a c c u r a c y s c o r e

p r i n t ( ” Confusion M a t r i x : ” )

print ( confusion matrix ( Y test , y pred ) )

p r i n t ( ” \ n Accuracy : ” , a c c u r a c y s c o r e ( Y t e s t , y p r e d ) )

plt . xlabel ( ’ Recall ’ )

plt . ylabel ( ’ Precision ’)

p l t . t i t l e ( ’ P r e c i s i o n − R e c a l l Curve wit h T h r e s h o l d ’ )

p l t . show ( )

# C a l c u l a t i n g t h e PR−AUC s c o r e wit h t h r e s h o l d

from s k l e a r n . m e t r i c s i m p o r t a v e r a g e p r e c i s i o n s c o r e

36
p r i n t ( ” \nPR−AUC Score wi t h T h r e s h o l d : ” , a v e r a g e p r e c i s i o n s c o r e ( Y t e s t , y p r
ed ) )

# C a l c u l a t i n g t h e AUC−PR s c o r e wit h t h r e s h o l d

from s k l e a r n . m e t r i c s i m p o r t auc

p r i n t ( ” \nAUC−PR Score wit h T h r e s h o l d : ” , auc ( r e c a l l , p r e c i s i o n ) )

# C a l c u l a t i n g t h e a r e a under t h e ROC c u r v e wi t h t h r e s h o l d

from s k l e a r n . m e t r i c s i m p o r t r o c a u c s c o r e

p r i n t ( ” \ n Area under t h e ROC c u r v e wi t h T h r e s h o l d : ” , r o c a u c s c o r e ( Y t e s t , y

pred ) )

# C a l c u l a t i n g t h e a r e a under t h e PR c u r v e wi t h t h r e s h o l d

from s k l e a r n . m e t r i c s i m p o r t a v e r a g e p r e c i s i o n s c o r e

p r i n t ( ” \ n Area under t h e PR c u r v e wi t h T h r e s h o l d : ” , a v e r a g e p r e c i s i o n s c o r e ( Y
test , y pred ) )

37
8.2 SAMPLE SCREENSHOTS:

38
39