Enhancing Fraud Detection in SWIFT Financial Systems Through Ontology-Based Knowledge Integration and Graph-Driven Analysis

Fraud detection within financial institutions presents a multifaceted challenge, necessitating robust tools for both the prevention and identification of fraudulent activities. In this article, we detail our methodology for detecting fraud within the SWIFT network, leveraging a domain-specific ontology. Our approach unfolds in two primary phases. Initially, we construct the OntoFiC ontology, tailored specifically for modeling SWIFT transactions and associated actors. This ontology is then populated with a real dataset. Following this foundational step, we employ rules-based reasoning, crafting SWRL (Semantic Web Rule Language) rules that correspond to various fraud scenarios. Subsequently, we construct a comprehensive graph representation to detect fraudulent customers and transactions, and we visualize the results of our fraud detection efforts. To assess the efficacy of our approach, we compare the performance of four anomaly graph detection algorithms (DOMINANT, GADNR, ONE and CoLA) with Isolation Forest algorithm. The results indicate that graph-based detection algorithms is better than traditional methods. Isolation Forest, while competitive in precision for legitimate accounts, demonstrated lower recall, suggesting it may miss some legitimate activities. These findings suggest that leveraging graph-based methodologies can enhance the detection of complex fraud patterns within financial transactions.