In 2018, criminal proceeds were estimated at 3.6% of global GDP with 2.7% successfully laundered. At more than a trillion dollars, the existence of this hidden economy enables criminal activity, facilitates government corruption, and suppresses the tax bases of developing nations. Node classification of transaction graphs as method of money laundering identification has been preliminarily explored by researchers, however, their results have been less than promising. This paper aims to improve the state-of-the-art of graph machine learning on neural networks, both in terms of accuracy and robustness. Our preliminary approach develops a prediction representation that incorporates the relative successes of random forest for predicting classifications for individual nodes and the network structure of a transaction graph dataset.

“I study computer science and economics because I am interested in understanding how technology can fight harmful structures that degrade our society. My SuperUrop research is an extension of that interest.”