MultiChannel and Multiple Device Financial Frauds Detection

Financial fraud has become a severe concern in the financial industry, affecting individuals and private and government organizations across the globe.Recently, Graph Neural Networks (GNN) have emerged as a promising tool for addressing these challenges.However, the effectiveness of these models remains uncertain due to the unrealistic nature of existing data.Traditional datasets available for financial research are often constrained by privacy concerns, incomplete labeling, and restricted access, hindering the development of effective detection models.To address these challenges, we first develop a synthetic dataset generator that mimics real-world financial transactions.The generator takes into account transaction patterns, multi-channel devices, and multiple accounts.Additionally, the generated dataset is simulated by extracting insights into fraudulent transaction patterns.We also leveraged the recently released IBM AML Sim world dataset, which provides valuable properties for financial fraud research.Second, we have used Deeper GCN model as it enhances fraud detection by capturing complex patterns across multiple GCN layers, detecting abnormal behaviors across accounts and devices.Dynamic edge updates refine transaction features, tracking evolving fraud patterns, while residual connections stabilize training, allowing deeper analysis in large transaction networks.We implemented our proposed model on the IBM AML dataset and observed a 1-2% improvement in results compared to existing benchmarks.