Machine Learning Approach to Detect Fraudulent Banking Transactions

Table of Contents

1 Introduction

2 Objective

3 Literature Review

3.1 Related Work

3.2 Machine learning approaches

3.2.1 Logistic Regression

3.2.2 Decision Tree

3.2.3 Working of Decision Tree

3.2.4 Random Forest

3.2.5 Support Vector Machines (SVM)

3.2.6 K-Nearest Neighbours (KNN)

3.2.7 Gradient Boosted Trees

3.2.8 Research Method Data Challenges

3.3 Recent Fraud Cases

3.4 CRISP-DM Model

3.4.1 Business Understanding

3.4.2 Data Understanding

3.4.3 Data Preparation

3.4.4 Data Modelling

3.4.5 Model evaluation

3.4.6 Model Deployment

4 Methodology and Case Study

4.1 Banking Theory

4.2 Data Description

4.3 Data Preparation

4.3.1 Scaling the data

4.3.2 Missing values handling

4.3.3 Dropping NA

4.3.4 Data encoding

4.4 Data Visualisation

4.4.1 Univariate Analysis

4.4.2 Histograms

4.4.3 Boxplot

4.4.4 Bivariate analysis

4.4.5 Correlation

4.4.6 Summary from EDA

4.5 Feature Selection

4.5.1 ANOVA

4.6 Model Comparison and Results

4.6.1 Logistic Regression

4.6.2 Decision Tree

4.6.3 Random Forest

4.6.4 XGBBoost

4.6.5 GradientBoosting

4.6.6 LGBMclassifier

4.7 Classification Evaluation Metrics

4.7.1 Confusion Matrix

4.7.2 Precision

4.7.3 Recall

4.7.4 F1 score

4.7.5 AUC-ROC

4.7.6 Receiver operating characteristic (ROC) Curve

4.7.7 Accuracy

4.7.8 Imbalanced Data

4.8 Possible next steps

5 Summary and Conclusion

Research Objective and Scope

This thesis aims to develop a robust machine learning-based system to detect fraudulent credit card transactions by analyzing historical payment data. The study addresses challenges such as dataset imbalance and provides a systematic comparison of various classification models to optimize fraud detection performance.

• Comparison of machine learning techniques for fraud identification.
• Analysis of credit card payment behavior to assess default risk.
• Implementation of oversampling/undersampling to handle unbalanced datasets.
• Evaluation of classification models using metrics like Accuracy, AUC-ROC, and F1 scores.

Excerpt from the Book

Summary of Chapters

1 Introduction: Provides an overview of the rising problem of fraud in the banking sector and highlights the necessity for advanced predictive models.

2 Objective: Outlines the research goals, including the identification of the best-performing fraud detection models and the handling of unbalanced data.

3 Literature Review: Discusses existing machine learning approaches and the CRISP-DM methodology used for fraud detection.

4 Methodology and Case Study: Describes the dataset, data preparation, feature selection, and the comparative results of classification algorithms.

5 Summary and Conclusion: Summarizes the key findings and provides recommendations for integrating these models into bank risk management systems.

Keywords

Machine Learning, Fraud Detection, Credit Card Default, CRISP-DM, Logistic Regression, Decision Tree, Random Forest, XGBoost, LGBMclassifier, Data Mining, Fraud Prevention, Risk Management, Classification, AUC-ROC, Imbalanced Data

Frequently Asked Questions

What is the primary focus of this research?

The research focuses on utilizing machine learning algorithms to detect fraudulent banking and credit card transactions to minimize financial losses.

What are the core thematic areas covered?

The core areas include fraud detection theory, the application of various supervised machine learning models, exploratory data analysis of payment behavior, and strategies for managing imbalanced datasets.

What is the main objective of this study?

The objective is to find a high-performing classification model that efficiently identifies potential fraudulent transactions and default payments in complex, real-world datasets.

Which scientific methods are employed?

The study employs the CRISP-DM (Cross-Industry Standard Process for Data Mining) model to structure the research, alongside algorithms like Logistic Regression, Random Forest, SVM, XGBoost, and LGBMclassifier.

What topics are discussed in the main body?

The main body detail the literature review, data description, data cleaning and encoding, Exploratory Data Analysis (EDA), feature selection techniques like ANOVA, and a comparative performance analysis of various ML models.

Which keywords characterize this work?

The work is characterized by terms such as Machine Learning, Fraud Detection, Credit Card Default, Data Mining, Classification, and Risk Management.

How does the study handle the imbalance in credit card data?

The research addresses imbalanced data using techniques like SMOTE (Synthetic Minority Oversampling Technique), undersampling, and oversampling to enhance the detection of the minority fraudulent class.

What makes the LGBMclassifier stand out in this study?

The LGBMclassifier achieved the best results in terms of both accuracy and AUC scores when compared to other supervised learning models like Decision Trees and Random Forests.