<![CDATA[The rapid growth of digital payment systems has increased the complexity of financial transactions, making credit card fraud detection more challenging, particularly due to evolving fraud patterns and highly imbalanced datasets. Conventional machine learning approaches often struggle to capture temporal dependencies and adapt to new fraud behaviors, while centralized data processing raises privacy concerns. This study proposes a hybrid fraud detection framework that integrates Bidirectional Long Short-Term Memory (BiLSTM), Autoencoder, and Federated Learning to improve detection performance while preserving data confidentiality. The BiLSTM component models sequential transaction behavior from both forward and backward directions, while the autoencoder identifies anomalies based on reconstruction errors. Federated Learning enables collaborative model training across multiple institutions without sharing sensitive data. Experimental evaluation using benchmark datasets shows that the proposed model achieves high classification performance, with improved precision, recall, and overall stability compared to traditional and standalone deep learning models. The framework effectively handles class imbalance and detects both known and emerging fraud patterns. This study contributes a scalable and privacy-preserving solution for real-world fraud detection, supporting secure collaboration and enhancing model generalization in distributed financial environments.]]>
Copyrights © 2026
