Fraud Detection in Blockchains using Machine Learning

Crypto crimes peaked in 2021, and recent years have shown that blockchain is not going to die but will stay with us. In addition, blockchain transaction throughputs are expected to rise, which will make the manual investigation of crimes next to impossible in the future. We need effective and automated solutions to monitor and detect fraudulent activities happening on blockchains. In this paper, we propose a system to detect the blacklisted addresses in the Ethereum blockchain. First, we collected Ethereum blockchain transaction data and blacklisted addresses. Then, we constructed the transaction graph of Ethereum and extracted features of addresses, including some global features like pagerank. Finally, we trained the models of standard machine learning algorithms and predicted the class of the addresses. Our results show that with more than 97% accuracy we can predict blacklisted addresses.