aiInternal Project
Fraud Detection System
Restaurant transaction fraud detection system achieving 97% accuracy using synthetic data generation and automated pipeline orchestration.
Developed at ATP
Senior Data Scientist
2022-2023
Internal project - no public access
Screenshots
About This Project
Built a high-accuracy fraud detection system for restaurant transactions that identifies fraudulent activities with 97% accuracy. The system leverages synthetic data generation techniques to handle class imbalance in fraud datasets and uses Apache Airflow for efficient process orchestration and automated pipeline management.
Key Features
- Real-time transaction fraud scoring and classification
- Synthetic data generation for handling imbalanced datasets
- Apache Airflow orchestrated data pipelines
- Automated model retraining and monitoring
- Integration with restaurant POS systems
- Detailed fraud analysis reports and dashboards
Challenges & Solutions
- Handling highly imbalanced datasets with very few fraud examples
- Generating realistic synthetic fraud data for model training
- Building a pipeline that processes transactions in near real-time
- Minimizing false positives to avoid blocking legitimate transactions
Results & Impact
- 97% accuracy in fraud detection
- Significant reduction in fraudulent transaction losses
- Automated end-to-end pipeline with Apache Airflow
Technologies Used
PythonScikit-LearnApache AirflowPandasNumPyDockerSQL
Project Details
- Category
- ai
- My Role
- Senior Data Scientist
- Duration
- 6 months
- Year
- 2022-2023
- Company
- ATP
- Status
- Internal / Private
Internal Project
This project was developed for internal use at ATP. Source code and live demo are not publicly available due to confidentiality.