Credit Card Fraud Detection using SOM and Random Forest

This project demonstrates a method to detect fraudulent credit card transactions using Self-Organizing Maps (SOM) for feature extraction and a Random Forest classifier for classification. The dataset used is the "Credit Card Fraud Detection" dataset from Kaggle.

Project Overview

Fraud detection is a critical task in the financial sector, where identifying fraudulent transactions can save significant amounts of money and enhance customer trust. This project leverages SOM for unsupervised feature extraction, followed by a Random Forest classifier to achieve high accuracy in detecting fraudulent transactions.

Dataset

The dataset used for this project is from Kaggle, and it contains anonymized credit card transactions labeled as either fraudulent or genuine.

• Credit Card Fraud Detection Dataset by MLG-ULB

Project Structure

• creditcard.csv: The dataset file.
• train_and_save_models.py: Script for training the models and saving them.
• predict_new_transaction.py: Script for loading the models and predicting new transactions.
• README.md: Project documentation.
• potential_frauds.csv: The output file containing potential fraudulent transactions identified by the model.

Requirements

The following Python libraries are required to run the project:

• pandas
• numpy
• scikit-learn
• minisom
• joblib

You can install these libraries using pip:

pip install pandas numpy scikit-learn minisom joblib

Usage

Training and Saving Models

1. Clone the repository:

git clone https://github.com/Taufiq-ML/credit-card-fraud-detection.git
cd credit-card-fraud-detection

2. Download the dataset:

   Place the creditcard.csv file in the root directory of the project. You can download the dataset from Kaggle.

3. Run the training script:

python train_and_save_models.py

This script will train the models and save them as scaler.pkl, som.pkl, and classifier.pkl.

Predicting New Transactions

1. Run the prediction script:

python predict_new_transaction.py

This script will load the saved models and predict whether the provided transaction is fraudulent or not.

Example Usage

Modify the new_user_transaction dictionary in predict_new_transaction.py with the actual transaction data you want to check. Here is an example:

new_user_transaction = pd.DataFrame([{
    'Time': 100000, 'V1': -1.359807, 'V2': -0.072781, 'V3': 2.536346, 'V4': 1.378155, 'V5': -0.338321,
    'V6': 0.462388, 'V7': 0.239599, 'V8': 0.098698, 'V9': 0.363787, 'V10': 0.090794, 'V11': -0.551600,
    'V12': -0.617801, 'V13': -0.991390, 'V14': -0.311169, 'V15': 1.468177, 'V16': -0.470400, 'V17': 0.207971,
    'V18': 0.025791, 'V19': 0.403993, 'V20': 0.251412, 'V21': -0.018307, 'V22': 0.277838, 'V23': -0.110474,
    'V24': 0.066928, 'V25': 0.128539, 'V26': -0.189115, 'V27': 0.133558, 'V28': -0.021053, 'Amount': 149.62
}])

Methodology

1. Data Preprocessing:

   • Load and standardize the dataset.

2. Self-Organizing Maps (SOM):

   • Train SOM to map high-dimensional data to a 2D grid.

3. Feature Extraction:

   • Use SOM to extract features by mapping each transaction to its corresponding winning neuron.

4. Random Forest Classifier:

   • Train a Random Forest classifier on the SOM features.
   • Evaluate the classifier on a test set.
   • Save the transactions predicted as fraudulent.

Results

The combination of SOM for feature extraction and a Random Forest classifier achieves high accuracy in detecting fraudulent transactions. The results are summarized in the classification report and accuracy score output by the script.

runfile('C:/Users/UseR/Desktop/Fraud/train_and_save_models.py', wdir='C:/Users/UseR/Desktop/Fraud') SOM training completed. Accuracy: 1.00 Classification Report: precision recall f1-score support

       0       1.00      1.00      1.00     85307
       1       0.79      0.36      0.49       136

accuracy                           1.00     85443

macro avg 0.89 0.68 0.75 85443 weighted avg 1.00 1.00 1.00 85443

Potential fraudulent transactions saved to 'potential_frauds.csv'.

runfile('C:/Users/UseR/Desktop/Fraud/predict_new_transaction.py', wdir='C:/Users/UseR/Desktop/Fraud') The transaction is: Not Fraudulent

Contributing

Contributions are welcome! Please open an issue or submit a pull request if you have suggestions for improvements or new features.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgements

• The dataset used in this project is provided by MLG-ULB on Kaggle.
• The SOM implementation is based on the minisom library.