PyTorch Geometric - Fraud Detection - 05/08/2024 03:58 EDT
Budget: €30 – €250 EUR
I'm in need of a PyTorch Geometric expert to develop a credit card fraud detection system using a graph neural network. The key components of the project include:
- Using 3 CSV files (transactions, merchants, cards) to create a heterogeneous graph
- Leveraging the graph neural network to enhance fraud detection
- Addressing the imbalanced classification problem typical in fraud detection
Key Requirements:
- Strong understanding of PyTorch Geometric framework
- Experience in credit card fraud detection or similar anomaly detection projects
- Familiarity with handling and processing heterogeneous data
- Proven track record in addressing imbalanced classification problems
Your work will involve developing an efficient and effective fraud detection system, ensuring high accuracy and precision. You'll be evaluated on the performance of the system using precision, recall, F1 score, and any additional metrics you suggest. If possible, you can use a dynamic GNN, which makes use of the time aspect (see paper DGNN) In addition, I would like to have a benchmarking (traditional) method which does not exploit the graph structure (like XGBoost or Random Forests, or other suggestions)
If you're an expert in PyTorch Geometric and have a strong background in fraud detection, I'd be thrilled to discuss this project with you.
- Using 3 CSV files (transactions, merchants, cards) to create a heterogeneous graph
- Leveraging the graph neural network to enhance fraud detection
- Addressing the imbalanced classification problem typical in fraud detection
Key Requirements:
- Strong understanding of PyTorch Geometric framework
- Experience in credit card fraud detection or similar anomaly detection projects
- Familiarity with handling and processing heterogeneous data
- Proven track record in addressing imbalanced classification problems
Your work will involve developing an efficient and effective fraud detection system, ensuring high accuracy and precision. You'll be evaluated on the performance of the system using precision, recall, F1 score, and any additional metrics you suggest. If possible, you can use a dynamic GNN, which makes use of the time aspect (see paper DGNN) In addition, I would like to have a benchmarking (traditional) method which does not exploit the graph structure (like XGBoost or Random Forests, or other suggestions)
If you're an expert in PyTorch Geometric and have a strong background in fraud detection, I'd be thrilled to discuss this project with you.