FraudNetGuard

Inspiration

Financial fraud continues to rise , costing banks and individuals billions in losses annually. We were inspired personally, when our friend'd dad got duped by a financial fraudster and the bank was helpless, as it was not able to detech the fraudster in it's system, we then ought to make a reliable and secured fraud detection system with our technical abilities.

What it does

We use the power of graphs and data visulazation bringing out the intrinsic network of frauds. By leveraging a type of neuraal network called Graph Neural Networks (GNN), which can find the hidden and complex fraud networks and alerting the banks to flag and rectify the fraud and act before a harm is caused. We also deepened our understanding of privacy-preserving machine learning with the implementation of SMPC (Secure multi-party computation) , which keeps the user data safe.

How we built it

We have a prototype based system in our hands and we will be working and completing the project within a period of time. The way we will be completing is: Data Processing - We structured financial transactions into a graph database (Neo4j), modeling account relationships. Graph Neural Network (GNN) Model - We implemented GNN-based fraud detection to analyze transaction patterns and detect fraud rings. Secure Multi-Party Computation (SMPC) - We encrypted transaction data using Homomorphic Encryption (HE) to enable secure fraud detection. API Development - We developed a FastAPI backend to process encrypted transactions and interact with the fraud detection model. Visualization & Alerts - We built a React.js frontend to display fraud networks and alert banks of high-risk transactions

Challenges we ran into

Computational Overhead: Training GNN models on large transaction graphs required optimization for real-time fraud detection. Encryption Complexity: Homomorphic encryption computations were costly, so we optimized encryption schemes to balance security and efficiency. Regulatory Constraints: Ensuring compliance with GDPR & AML regulations while performing cross-bank fraud analysis was challenging.

Accomplishments that we're proud of

Despite these challenges, we successfully built a prototype that enhances fraud detection while protecting financial privacy

What we learned

we explored cutting-edge technologies like Neo4j for graph-based fraud analysis, PyTorch Geometric for GNN implementation, and SMPC for user data encryption.