| dc.description.abstract | With the emergence of cryptocurrencies and blockchain technology, the paradigm of the structure of data storing and distribution has completely changed. And while the central goal of Bitcoin's 2008 whitepaper was to create internet-based peer-to-peer money without a central third party, there are some unforeseen issues that come with having a completely decentralized ledger. Money laundering is possible by moving illicit funds through hundreds of wallets before depositing the funds and cashing out with a crypto exchange. And with these methods of money laundering becoming more advanced over the decades, the advent of cryptocurrency means a new venue for criminals to carry out more malicious schemes. Anti-money laundering techniques have given way to software within the world of technology that is becoming heavily used by financial institutions to analyze and detect suspicious data. The Elliptic Data Set, which maps Bitcoin transactions to real entities categorized as either from a licit or illicit group, is utilized in this study. It is known as the largest data set that is publicly available by any cryptocurrency and is made up of over 200,000 nodes and 230,000 edges. To aid in the fight against money laundering schemes, we discuss and analyze areas of machine learning that can benefit AML. Visualization of the graph constructed from the data set is examined to show any promising patterns in the location of licit vs illicit nodes. Furthermore, various binary classification models are used on the Elliptic data set in analyzing performance of prediction of illicit nodes within the network. The classification models include Logistic Regression, K-Nearest Neighbors, Decision Trees, Multilayer Perceptron, Random Forest, and Graph Attention Networks. The results of running these models on the data set provided insight into how nodes in the blockchain network are structured while also demonstrating innovate ways that machine learning can be leveraged within the AML industry. | |
