Summary: The paper compares several supervised learning techniques to detect money laundering in bitcoins, using a company labelled dataset that they provide publicly. This is a tutorial presented at a KDD 2019 workshop: Strengths: I particularly loved the direct, clear and convincing writing style of the paper. Also, the authors (seem to) make their data publicly available, and it is a very interesting dataset (in particular thanks to the labelling). Weaknesses: The used data suffers from several limitations: it is very partial (approx 200 K transactions, while 400 M are available), and the labelling is not presented in details. More globally, several aspects of the work are not detailed, and many choices seem arbitrary. But the authors themselves state that this is "early experimental results" only, aimed at "inspiring others to work on this societally important challenge".

Please consider to register or login to comment on the paper.