Behavior-Driven Simulation of Cryptocurrency Transactions for Anti-Money-Laundering Model Training
A method for overcoming the scarcity of labeled data in cryptocurrency anti-money-laundering (AML) by simulating money-laundering-like transactions instead of relying on a single fixed benchmark. It embeds the behaviors of the distinct entities in the crypto ecosystem (launderers, exchanges, ordinary users) to generate diverse, configurable, on-demand synthetic transaction datasets, addressing class imbalance and letting detection models be trained on realistic, scenario-tailored laundering patterns.
2501.00757
Cryptocurrencies are widely abused for money laundering, but training machine-learning detectors is hampered by scarce, static, class-imbalanced transaction datasets that cannot be customized to newl…