Synthetic Identity Detection Using Graph Neural Networks

Synthetic identity fraud has become one of the most significant problems for financial institutions and online services. This is because synthetic identities can use a combination of real and false information, and through non-linear relationships and changing patterns, are difficult to identify using conventional procedures. This paper will explore Graph Neural Networks (GNNs) in the context of synthetic identity detection, which lies in their potential to capture the structure underlying relational data and learn complex network representations. As an adaptive way of thinking, GNNs utilise the graph structure to represent the identities of users, transaction histories, and social ties, taking into account dependencies and abnormal patterns that are not captured by traditional machine learning algorithms. We present our solution that consists of feature engineering, graph construction, and classification using GNNs, which shows its efficiency, built on benchmark datasets. There will be an increase in detection, a decrease in false positives, and an increase in interpretability. This paper proposes a contribution to the field of fraud detection by combining deep learning methods with graph-based learning to achieve a scalable and dynamic algorithm for classifying synthetic names in dynamic, networked digital environments