New Machine Learning Methods Uncover Criminal Activity in the Bitcoin Blockchain

Researchers from Elliptic, in collaboration with the MIT-IBM Watson AI Lab, have developed new machine learning techniques that can detect criminal activity on the Bitcoin blockchain, including money laundering and transfers to suspicious wallets.

The study analyzed a 26-gigabyte dataset containing 122,000 labeled subgraphs within the blockchain, which includes 49 million nodes and 196 million transactions. The dataset, named Elliptic2 by the researchers, enabled them to identify connections between wallets and transactions associated with illegal activities on the blockchain.

It’s clear that “Elliptic2” is a continuation of the “Elliptic1” research, which was originally published in July 2019. The project’s goal is to combat financial crime using machine learning technologies, specifically graph convolutional neural networks (GCN).

Tom Robinson, Chief Scientist and Co-founder of Elliptic, explained that using machine learning at the subgraph level allows for the prediction of whether certain crypto transactions are proceeds from criminal activity. This approach differs from traditional analysis methods, which focus on tracking the activity of known illegal crypto wallets.

The study applied three subgraph classification methods: GNN-Seg, Sub2Vec, and GLASS. These methods helped identify numerous crypto exchange accounts potentially involved in illicit activities.

The researchers also identified various cryptocurrency laundering patterns, including the so-called “Peeling Chain.” Future research will focus on improving the accuracy and detail of these analytical methods, as well as expanding their application to other blockchains.