FraudHawk is an advanced anomaly detection project designed to monitor user transactions and identify fraudulent activities in real-time. By leveraging machine learning techniques such as Autoencoders and Isolation Forest, FraudHawk provides a robust framework for ensuring transaction integrity, enhancing user trust, and minimizing financial losses due to fraud.
- Real-time Transaction Monitoring: Detects and alerts users about potentially fraudulent transactions as they occur.
- Flexible Input Format: Supports CSV file uploads with transaction details including timestamps, amounts, categories, and transaction types (online/onsite).
- Anomaly Detection Models: Utilizes both Autoencoders and Isolation Forests for identifying anomalies, providing a comparative analysis of both approaches.
- User-friendly Dashboard: Visualizes transaction histories and highlights anomalies using interactive plots.
1. Autoencoders:
- Pros:
- Effective for high-dimensional data.
- Learns complex patterns and reconstructs inputs, highlighting anomalies based on reconstruction error.
- Can handle various types of data seamlessly.
- Cons:
- Requires careful tuning of architecture and hyperparameters.
- May need substantial computational resources for training.
2. Isolation Forest:
- Pros:
- Efficient in terms of both speed and memory usage.
- Well-suited for detecting anomalies in large datasets with high-dimensional features.
- Simple to implement and interpret.
- Cons:
- Assumes that anomalies are few and differ significantly from the normal observations.
- Less effective if the data has a uniform distribution.
Upload your transaction data in CSV format. Choose to start transaction simulation to monitor for anomalies. View real-time alerts and visualizations on the dashboard.
To set up the FraudHawk project locally, follow these steps:
git clone https://github.com/HrithikRai/FraudHawk.git
cd FraudHawk
pip install -r requirements.txt
uvicorn backend:app --reload
streamlit run app.py