The Aondover Peter is an M.Sc. student in Applied Mathematics and Computer Science at the Moscow Institute of Physics and Technology (MIPT), Russia, specializing in Advanced Combinatorics. He obtained a First Class Bachelor's degree in Mathematics from the Federal University Wukari, Nigeria, graduating as the overall best student of the university, faculty, and department. His research interests include Graph Neural Networks, Explainable Artificial Intelligence (XAI), Graph Theory, Financial Fraud Detection, Fluid Mechanics, and Mathematical Modeling. His current research focuses on developing explainable graph-based artificial intelligence models for financial fraud detection. He has extensive teaching experience in mathematics and has received several academic awards, including the Open Doors Scholarship and multiple Best Graduating Student awards.

Topic: Explainable AI for Financial Fraud Detection: A Graph Based Approach   The increasing sophistication and networked nature of financial fraud present a significant challenge to global financial systems. While Graph Neural Networks (GNNs) have demonstrated superior performance in detecting complex, relational fraud patterns, their inherent ``black-box" nature limits their practical adoption, as financial institutions require transparent and actionable explanations to trust and act upon AI-generated alerts. This study addresses the critical need for explainability in AI-driven financial fraud detection by developing, implementing, and rigorously evaluating a framework that combines a high-performance Graph Attention Network (GAT) with a comparative analysis of Graph Explainable AI (GXAI) techniques. Using the Elliptic Bitcoin Dataset (203,769 nodes, 234,355 edges), a robust GAT fraud detection model was established. Three prominent GXAI methods-Gradient-Based, Statistical PGExplainer, and Ablation-Based-were then applied and evaluated using a multi-faceted framework focusing on fidelity (faithfulness to model reasoning), sparsity (conciseness), and stability (consistency).The experimental results reveal a clear trade-off: the Gradient-Based method excelled in sparsity, the Ablation-Based method in stability, but the Statistical PGExplainer achieved the highest fidelity.Given the paramount importance of explanation accuracy in high-stakes fraud investigation and regulatory compliance, this study identifies the Statistical PGExplainer as the most suitable GXAI method for operational deployment in financial contexts. The work provides a foundational benchmark and practical framework for integrating explainability into AI-driven fraud detection systems, bridging the gap between theoretical GXAI research and the practical needs of thefinancial industry. Future directions include developing hybrid explanation methods and expanding evaluations to dynamic, temporal financial graphs.