Machine Learning Algorithms to Detect Illicit Accounts on Ethereum Blockchain

Abstract

The rapid growth and psudonomity inherent in blockchain technology such as in Bitcoin and Ethereum has marred its original intent to reduce dependant on centralised system, but created an avenue for illicit activities, including fraud, phishing, scams, etc. This undermines the reputation of blockchain network, giving rise to the need to identify these illicit activities within the blockchain network. This current work tackles this crucial problem by investigating and implementing six machine learning algorithms with a particular emphasis on striking a balance between accuracy, precision and recall. The novelty of the work lies in the utilising of the synthetic minority over-sampling technique to handle data imbalance. Thus, increasing the accuracy of the light gradient boosting machine classifier to 98.4%. The outcome of this work holds great potential for enhancing the security and credibility of blockchain ecosystems paving the way for a more secure and dependable digital future in the age of decentralised and trustless systems.