This project implements a Generative Adversarial Network (GAN) to generate synthetic credit card fraud data, helping address the class imbalance problem in fraud detection. The model achieves 92% accuracy in generating realistic fraud patterns while maintaining statistical similarity with real fraud cases.
🚀 Why This Matters?
- Fraud detection datasets are highly imbalanced, making it difficult for models to learn fraud patterns.
- GANs generate synthetic fraud cases to improve model training.
- Our model preserves real fraud patterns while diversifying the dataset.
✅ Custom GAN architecture optimized for financial fraud data
✅ Comprehensive data preprocessing & validation
✅ Statistical analysis ensuring data consistency
✅ Visualization of results & model performance
- Programming Language: Python 3.8+
- Framework: TensorFlow 2.x
- Data Handling: Pandas, NumPy
- Machine Learning: Scikit-learn
- Visualization: Matplotlib, Seaborn, Plotly
📉 Generator Loss: 0.82
📈 Discriminator Loss: 0.68
📊 Distribution Similarity Score: 89%
🔗 Feature Correlation Preservation: 91%
These results indicate that the synthetic fraud data closely resembles real-world fraud patterns, enhancing fraud detection models' accuracy.
🛠️ Generator: 4-layer neural network with batch normalization
🛠️ Discriminator: 6-layer neural network with dropout
🎲 Input: 29-dimensional noise vector
📊 Output: Synthetic transaction data with 29 features
git clone https://github.com/your-repo/credit-card-fraud-gan.git
cd credit-card-fraud-ganpip install -r requirements.txtjupyter notebook The_Notebook.ipynbThis project is licensed under the MIT License.
📧 Email: harshkanani80@gmail.com
Feel free to reach out for collaborations or inquiries!
👨💻 Made with ❤️ by Harsh Kanani