Deep Fraud Detection in Cloud-Based Banking Systems Using Recurrent Neural Networks and Graph Convolutional Networks

Abstract

Fraudulent rule detection in a cloud banking system has become a highly critical problem due to not only the complex sophisticated cyber era and fraud but also the complexity introduced by the conventional rule-based approaches. Such approaches tend to fail to detect sophisticated fraud patterns, thereby generating huge numbers of false positives and rendering the detection process inefficient. Thus, this paper presents a model incorporated into hybrid deep learning for enhancing the accuracy of fraud detection by combining Recurrent Neural Networks (RNNs) and Graph Convolutional Networks (GCNs). The RNNs learn temporal data in the sequence of transactions and the GCNs learn the relationships in the financial network. The model was tested with the PaySim dataset and had an accuracy of 96.5% precision and recall of 95.2% and 94.8%, and an AUC-ROC of 98.2%, thereby establishing its insensitivity to fraud classification. The results indicate that sequential and graphic performance enhances overall fraud detection efficiency, reducing false positives and negatives. This study will promote intelligent fraud detection systems that ensure security and reliability in all electronic financial transactions