<![CDATA[Money laundering accounts for an estimated 2–5% of global GDP annually with scale intensified by digital ecosystems. Conventional AML systems using primarily rule-based and transactional patterns struggle to detect relational behaviors of financial crimes. This study introduces an integrated graph-analytic framework to detect structural laundering patterns using graph-derived metrics to neural network pipeline. The paper evaluates eccentricity, degree, closeness measures, and directionality of flow to distinguish laundering activities, supported by Welch’s t-test which confirms statistically significant differences across five of six metrics (p < 0.001). A Multi-Layer Perceptron (MLP) model is further applied to classify 17 typologies with ~80% accuracy. The key contribution of this research lies in demonstrating that financial crime typologies can be extracted from network topology itself instead of sole reliance on transactional features. By linking graph metrics with laundering behaviors including placement, layering, and integration patterns the study provides a scalable, network-aware approach to AML detection. Future work should focus on real-world validation and real-time classification pipelines using graph-neural inference.]]>
Copyrights © 2025
