Fraud Detection using Synthetic Data

Overview

This project aims to develop a machine learning model capable of detecting fraudulent transactions using synthetic data.

Data Exploration

Distribution of Transaction Amounts

This plot shows the distribution of transaction amounts. Most transactions fall within a specific range, with a few high-value transactions.

Count of Fraudulent vs Non-Fraudulent Transactions

This plot highlights the imbalance between fraudulent and non-fraudulent transactions, with fraudulent transactions being a small minority.

Data Preprocessing

Class Distribution Before and After Oversampling

These plots show the class distribution before and after oversampling. Oversampling helps balance the classes, which is crucial for training an effective model.

Model Evaluation

Model Performance Metrics

              precision    recall  f1-score   support

       False       1.00      0.98      0.99       189
        True       0.98      1.00      0.99       191

    accuracy                           0.99       380
   macro avg       0.99      0.99      0.99       380
weighted avg       0.99      0.99      0.99       380

The classification report provides precision, recall, and F1-score for both classes, showing that the model performs well on both non-fraudulent and fraudulent transactions.

ROC Curve

The ROC curve shows the trade-off between true positive rate and false positive rate for different threshold values. The area under the curve (AUC) is a measure of the model's ability to distinguish between classes, with a score of 1.0 indicating perfect performance.

Results

• Model achieved a ROC-AUC score of 1.0 after handling class imbalance.
• Visualizations and detailed analysis can be found in the Jupyter Notebook.

Usage

1. Clone the repository:

   git clone https://github.com/yourusername/fraud_detection_project.git

2. Navigate to the project directory and install dependencies:

   cd fraud_detection_project
   pip install -r requirements.txt

3. Run the Jupyter Notebook to see the full analysis:

   jupyter notebook notebooks/fraud_detection_notebook.ipynb

Author

Robert Grantham