FedAvg-DWA: A Novel Algorithm for Enhanced Fraud Detection in Federated Learning Environment

Abstract

The research on credit card fraud detection can protect consumers and banks from economic losses. The current research direction mainly focuses on improving the efficiency and accuracy of fraud detection using machine learning technology. This paper mainly studies how to use federated learning to improve the efficiency and effect of credit card fraud detection. The experiment uses the credit card transaction data of European cardholders and uses the Dirichlet distribution to randomly and proportionally allocate the sample data to 10 simulated banks. In response to the limited prediction accuracy of existing federated learning algorithms such as FedAvg (Federated Averaging) for imbalanced data, this paper proposes a FedAvg-DWA algorithm (Federated Averaging with Distance-based Weighted Aggregation). This algorithm considers the weight of small class samples and innovates in the model distance aspect of the aggregation strategy. The experiment achieved good results, providing a new method for dealing with the problem of class imbalance.