Overview
Crimes like money laundering and drug trafficking are only a small fraction of all transactions on cryptocurrency exchange networks, but they pose a serious threat. The decentralized and anonymous nature of these networks has made them a popular vehicle for bad actors to conduct and conceal illicit financial activity. These challenges have led law enforcement and financial agencies to develop machine learning methods to identify illicit cryptocurrency transactions at scale, but these AI tools need well-labeled data to effectively detect and stop such crimes—and the labeling process is expensive and time-consuming.
As authorities search for leaner ways to label data nodes and identify illicit crypto transactions, active machine learning offers a promising path forward. Active learning focuses on training AI tools to selectively choose nodes with the greatest uncertainty for focused investigation, then incorporates these findings into the AI’s process to iteratively improve accuracy. Building on previous NSPDI papers on detecting illicit financial activity, our researchers developed new active learning methods that incorporate network structure to improve AI performance and efficiency. In this paper, we describe how this work led to the creation of an algorithmic tool called TwoStepALWithCC, which uses active learning to more efficiently find hidden bad actors in financial networks even when data is incomplete, mislabeled, or immense.
Key Takeaways
- Because the active learning techniques used by the NSPDI focus human researchers and the AI algorithm on the most confusing (and therefore most informative) nodes, it helps the system learn faster and more accurately while reducing the total nodes that need labeling.
- NSDPI’s researchers’ proposed method of identifying illicit financial transactions using TwoStepALWithCC matched or surpassed existing techniques while using up to 22% fewer labeled samples.
- TwoStepALWithCC offers the intelligence community faster identification of suspicious nodes, more efficient allocation of investigative resources, and enhanced knowledge of vast financial datasets.