Improving Data Privacy and Performance in Collaborative Financial Fraud Detection with 3-Tiered Federated Learning with Differential Privacy

Amidst the rise of the new digital economy, combating financial fraud has become a critical challenge, requiring numerous collaborative efforts to detect fraudulent transactions in real-time. However, any collaborative solution needs to take privacy considerations into account while sharing sensitive data, which is a crucial hindrance to obtain optimal performance. This paper presents an in-depth analysis of the benefits of utilizing a 3-tiered Federated Learning architecture with DPFedAvg aggregation strategy as a collaborative alternative means to tackle financial fraud, allowing institutions to collaborate while maintaining privacy and further improving security. Additionally, we assess the effectiveness of this approach in detecting fraudulent transactions, comparing its performance to traditional methodologies. Our findings reveal a notable performance improvement using the Federated Learning model, surpassing conventional models.