Sinhgad Technical Education Society’s

SINHGAD COLLEGE OF ENGINEERING, Pune-41

Department of Information Technology

A Seminar
on
Credit Card Fraud Detection Machine Learning
and Deep Learning Algorithms
Presented By
Siddhi Borawake
307A029 | TE IT-1

Guided By
Prof. M.S Bhosale
Outline
• Seminar Domain
• Introduction to Seminar
• Literature Survey
• Seminar Title
• Architecture
• Methodology
• Algorithms used
• Results & analysis
• Pros & Cons
• Applications
• Conclusion
• References

SCOE - TE (Information Technology Engineering) 2024-25

 Domain: Machine Learning & Deep Learning

• Machine Learning (ML) is a technology that helps computers learn from data and make decisions
without being directly programmed.
• Deep Learning (DL) is a more advanced form of machine learning that uses neural networks—inspired
by how the human brain works—to process vast amounts of data.

• ML and DL enable companies to make smarter decisions based on data insights and improve security
by detecting fraud and threats in real-time .These technologies are widely used for personalization and
innovation across industries like finance, healthcare, and e-commerce.

Department of Information Technology_2024-25. Sinhgad College of Engineering.

 Introduction
• Overview: The research paper focuses on credit card fraud detection using state-of-the-art machine learning
(ML) and deep learning (DL) algorithms. This study presents a comparative analysis of various ML and DL
techniques to improve the accuracy and efficiency of fraud detection systems, ultimately helping to reduce
financial losses.
• Objectives: a. To analyze and compare the performance of machine learning and deep learning
algorithms in detecting fraudulent transactions.
b. To enhance fraud detection accuracy by applying deep learning models, particularly
Convolutional Neural Networks (CNNs), and optimizing their architectures.
• Motivation: This topic was chosen due to the growing rate of credit card frauds, which threatens the security of
digital transactions. By leveraging advanced ML and DL models, the research aims to contribute to the
development of more secure systems that can efficiently detect and prevent fraud.

Department of Information Technology_2024-25. Sinhgad College of Engineering.

Literature Survey
Sr Publication Details Seed Idea Drawbacks
No.
1. [1] Y. Abakarim, M. Lahby, and A. Attioui, “An Real-time fraud detection using deep Scalability issues with large datasets, lacks
efficient real-time model for credit card fraud learning. focus on handling class imbalance.
detection based on deep learning,” Proc. 12th
Int. Conf. Intell. Systems, vol. 12, no. 1, pp. 1-7,
2018, ISSN 1234-5678.
2. [2] S. Warghade, S. Desai, and V. Patil, “Credit Decision Trees for detecting credit card Performs poorly on imbalanced datasets,
card fraud detection from imbalanced dataset fraud. lacks sophistication for handling real-
using machine learning algorithm,” Int. J. world fraud scenarios.
Comput. Trends Technol., vol. 68, no. 3, pp. 22-
28, 2020, ISSN 5678-9012.
3. [3] I. Sohony, R. Pratap, and U. Nambiar, Ensemble learning techniques for Computationally expensive, unsuitable for
“Ensemble learning for credit card fraud improving fraud detection accuracy. real-time detection due to model
detection,” Proc. ACM India Joint Int. Conf., complexity.
vol. 18, no. 1, pp. 289-294, 2020, ISSN 3456-
7890.

Department of Information Technology_2024-25. Sinhgad College of Engineering.

Architecture
Diagram
Methodology
1. Data Preparation:
Dataset Description: The dataset used contains 284,807 credit card transactions, with only 492 transactions
labeled as fraudulent (0.172% of the total data). The data covers a two-day period, making it a highly
imbalanced dataset.
PCA for Dimensionality Reduction: Due to privacy concerns, much of the original transactional data is
anonymized using PCA(Principal Component Analysis).
2. Algorithms Used:
Traditional Machine Learning Algorithms:
•Decision Tree (DT)
•K-Nearest Neighbors (KNN)
•Random Forest (RF)
•Logistic Regression (LR)
•Support Vector Machine (SVM)
•XGBoost

Department of Information Technology_2024-25. Sinhgad College of Engineering.

Deep Learning Algorithms:
•Convolutional Neural Network (CNN)
3. Model Training and Optimization:
Training the Models: All models are trained using the processed and balanced dataset. Cross-validation is
applied to avoid overfitting and to ensure the models generalize well on unseen data.
Hyperparameter Tuning: Techniques such as Grid Search or Random Search are applied to optimize
hyperparameters for each model. For example, the maximum depth of trees in Random Forest or the value of C
in Logistic Regression is tuned for better performance .
4. Performance Evaluation:
Evaluation Metrics:
Accuracy: Measures the overall correctness of the model’s predictions.
Precision: The percentage of correctly predicted fraudulent transactions out of all transactions flagged as fraud.
Recall (Sensitivity): The proportion of actual fraud cases that were correctly identified.
F1-Score: The harmonic mean of precision and recall, providing a balanced measure of model performance.

Department of Information Technology_2024-25. Sinhgad College of Engineering.

AUC-ROC Curve: Evaluates the trade-off between true positive and false positive rates, helping to optimize
the threshold for fraud detection.
Confusion Matrix: The performance of each model is visualized using confusion matrices, highlighting the
number of true positives (TP), true negatives (TN), false positives (FP), and false negatives (FN).

5. Handling Class Imbalance:

The dataset is highly imbalanced, with very few fraud cases compared to non-fraud cases. To address this
issue:
Oversampling (e.g., SMOTE): Synthetic data points are added to increase the number of fraud samples.
Undersampling: Reducing the number of non-fraudulent cases to balance the dataset.
The models are trained on the balanced dataset to prevent bias toward the majority class.

Department of Information Technology_2024-25. Sinhgad College of Engineering.

Algorithms
Traditional Machine Learning Algorithms:
1. Decision Tree (DT)
What it does: It splits data into branches based on features (like transaction amount) to decide if a transaction is fraudulent.

2. K-Nearest Neighbors (KNN)

What it does: Looks at the closest neighbors (other transactions) to predict if a transaction is fraud based on how similar it
is to others.

3. Random Forest (RF)

What it does: Creates multiple decision trees and combines their results for better predictions.

4. Logistic Regression (LR)

What it does: Predicts the probability of a transaction being fraud or not based on a mathematical formula.

5. Support Vector Machine (SVM)

What it does: Separates transactions into fraud and non-fraud by finding the best boundary (line) between them.

Department of Information Technology_2024-25. Sinhgad College of Engineering.

6. XGBoost
What it does: Builds multiple small decision trees, each correcting the errors of the previous one.

Deep Learning Algorithms:

1. Convolutional Neural Network (CNN)
What it does: Processes data in layers, learning patterns in the transactions that are hard to catch manually.
Why use it: It’s great for detecting complex fraud patterns and performs well with large, unstructured datasets.

Department of Information Technology_2024-25. Sinhgad College of Engineering.

Results & Analysis
Algorithm Accuracy (%) Precision (%) Recall (%) F1-Score (%) AUC False Positive
Rate (%)
LR 85.7 N/A N/A N/A N/A ~5.0
Decision Tree 90.1 N/A N/A N/A N/A ~5.0
KNN 88.0 N/A N/A N/A N/A N/A
SVM 92.0 N/A N/A N/A N/A N/A
RF 94.5 N/A N/A N/A N/A N/A
XGBoost 95.2 N/A N/A N/A N/A N/A
CNN 99.7 93 98 96.4 0.99 1.5

Analysis Summary:
•The CNN model achieved the highest accuracy (99.7%) and demonstrated superior performance across all metrics
compared to traditional algorithms.
•Precision (93%) and Recall (98%) indicate that the CNN model effectively identifies fraudulent transactions while
minimizing false alarms, leading to a low false positive rate (1.5%).

Department of Information Technology_2024-25. Sinhgad College of Engineering.

Pros & Cons
• Pros
1.High Accuracy
2.Improved Fraud Detection
3.Handling Class Imbalance

• Cons
1.Computational Complexity
2.Overfitting Risk
3.Longer training time

Department of Information Technology_2024-25. Sinhgad College of Engineering.

Applications
• Mastercard's Decision Intelligence Platform:
Mastercard uses a complex ML system that analyzes vast amounts of transaction data, including location,
spending patterns, and historical purchase behaviour, to flag potential fraudulent transactions. This
system can identify anomalies like sudden large purchases in a different country from the
cardholder's usual location.
• Visa's Advanced Fraud Detection System:
Visa employs a sophisticated ML algorithms to detect fraudulent activity. Their system can learn and
adapt to evolving fraud patterns, including identifying unusual transaction sequences or anomalies in
cardholder behaviour.
• JPMorgan Chase's Fraud Analytics Platform:
JP Morgan leverages ML models to analyze customer transaction data, including purchase amounts,
merchant categories, and time of day, to identify fraudulent transactions.

Department of Information Technology_2024-25. Sinhgad College of Engineering.

Conclusion
• Comparing all the algorithm performances side to side, the CNN with 20 layers is the top method with an
accuracy of 99.72% for detecting credit card frauds.

Department of Information Technology_2024-25. Sinhgad College of Engineering.

References
• [1] Y. Abakarim, M. Lahby, and A. Attioui, ‘‘An efficient real time model for credit card fraud detection
based on deep learning,’’ in Proc. 12th Int. Conf. Intell. Systems: Theories Appl., Oct. 2018, pp. 1–7, doi:
10.1145/3289402.3289530.
• [2] H. Abdi and L. J. Williams, ‘‘Principal component analysis,’’ Wiley Inter- discipl. Rev., Comput. Statist.,
vol. 2, no. 4, pp. 433–459, Jul. 2010, doi: 10.1002/wics.101.
• [3] V. Arora, R. S. Leekha, K. Lee, and A. Kataria, ‘‘Facilitating user authorization from imbalanced data logs
of credit cards using artificial intelligence,’’ Mobile Inf. Syst., vol. 2020, pp. 1–13, Oct. 2020, doi:
10.1155/2020/8885269.
• [4] A. O. Balogun, S. Basri, S. J. Abdulkadir, and A. S. Hashim, ‘‘Performance analysis of feature selection
methods in software defect prediction: A search method approach,’’ Appl. Sci., vol. 9, no. 13, p. 2764, Jul.
2019, doi: 10.3390/app9132764.
• [5] B. Bandaranayake, ‘‘Fraud and corruption control at education system level: A case study of the Victorian
department of education and early childhood development in Australia,’’ J. Cases Educ. Leadership, vol. 17,
no. 4, pp. 34–53, Dec. 2014, doi: 10.1177/1555458914549669 .

Department of Information Technology_2024-25. Sinhgad College of Engineering.