Federated Learning-Based Credit Card Fraud Detection: A Comparative Analysis of Advanced Machine Learning Models
Because of the privacy concerns about the transaction data, it is essential not to leak it when training prediction models for credit card fraud analysis. Challenges for credit card fraud monitoring include highly imbalanced datasets and the need for advanced models to detect fraud patterns. This pa...
Saved in:
| Main Author: | |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
EDP Sciences
2025-01-01
|
| Series: | ITM Web of Conferences |
| Online Access: | https://www.itm-conferences.org/articles/itmconf/pdf/2025/01/itmconf_dai2024_01022.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Because of the privacy concerns about the transaction data, it is essential not to leak it when training prediction models for credit card fraud analysis. Challenges for credit card fraud monitoring include highly imbalanced datasets and the need for advanced models to detect fraud patterns. This paper introduced federated learning and discussed a few federated learning algorithms applied to the problem—these methods include Federated Graph Attention Network with Dilated Convolution Neural Network (FedGAT-DCNN), FedAvg with Convolutional Neural Network (CNN), and Federated Averaging with Distance-based Weighted Aggregation (FedAvg-DWA) with Random Forest (RF). Federated Averaging (FedAvg) aggregates data from local clients and then creates a global model; fedavg-dwa provides dynamic weight averaging, which enhances each client’s performance based on their data quality. The FedGAT-DCNN model improves accuracy by integrating GAT with Dilated Convolutions to catch spatial and temporal patterns in transaction data. FedGAT—DCNN performs best on highly imbalanced datasets with a high Area under the Receiver Operating Characteristic Curve (ROC-AUC) score. FedAVG-DWA provides the best performance in different clients’ systems. However, system heterogeneity, communication costs, and data imbalance remain critical. Oversampling techniques, model optimization, and reduced communication rounds were used to mitigate the issues. Therefore, federated learning’s ability can enhance credit card fraud sensing issues without privacy concerns. The paper highlights both the benefits and challenges of using federated learning in the domain. |
|---|---|
| ISSN: | 2271-2097 |