Fraud Analysis

Description

This project focuses on analyzing mobile money transactions to identify fraudulent activities using the PaySim dataset. The dataset is a synthetic simulation based on real financial logs from a multinational financial service provider operating across 14 countries.

Dataset Overview

The PaySim dataset contains a scaled-down version (1/4th of the original size) of mobile transaction logs. It provides data points such as transaction types, amounts, and whether a transaction was flagged as fraudulent.

Key Features:

• Synthetic Data: Simulated data ensures privacy while maintaining statistical integrity.
• Transaction Types: Includes multiple transaction categories.
• Target Variable: Flag indicating whether a transaction is fraudulent.

Methodology

Hypothesis Testing

Performed the following statistical tests to analyze transaction patterns and fraud detection:

• Chi-Square Test
• T-Test
• Mann-Whitney U Test
• ANOVA Test

Machine Learning Models

Implemented various models to predict fraudulent transactions:

• Logistic Regression
• Decision Tree Classifier
• Linear Discriminant Analysis
• Gaussian Naïve Bayes
• Support Vector Machine
• K-Nearest Neighbors Classifier

Handling Imbalanced Data

• Under-sampling: Applied to balance the dataset and improve model performance.

Tools and Libraries

• Visualization: Tableau
• Statistical Tests: Scipy, Statsmodels

Results and Insights

• Hypothesis Testing: Identified significant patterns and associations between transaction categories and fraudulence.
• Machine Learning Models: Demonstrated varying accuracy levels, with each model providing unique insights into fraud detection.

Visualization: Tableau

Link

Trello:

Link