The AMLNet framework introduces a novel approach to anti-money laundering (AML) research by providing a regulation-aware transaction generator and an ensemble detection pipeline. This dual-unit structure enables the generation of 1,090,173 synthetic transactions that effectively mimic various money laundering phases and typologies, ensuring approximately 0.16% of the transactions are laundering-positive. With a regulatory alignment reaching 75% based on AUSTRAC rules, the framework aims to bridge the gap between regulatory compliance and the need for effective synthetic data in AML studies.
The ensemble detection system boasts an impressive F1 score of 0.90, indicating its effectiveness in identifying potential laundering activities. This framework not only adapts to existing datasets like SynthAML but also emphasizes architectural generalizability across different paradigms of synthetic data generation. The implications of AMLNet are significant, as it promotes reproducible research practices within the AML domain, encouraging researchers to utilize the released dataset to advance their experiments and findings.
👉 Pročitaj original: arXiv AI Papers
