On the Combination of Machine Learning Techniques with Simulated Annealing to Identify Professional Accounts in Ethereum

• Júlia Almeida Valadares UFJF
• Saulo Moraes Villela UFJF
• Heder Soares Bernardino UFJF
• Alex Borges Vieira UFJF

Resumo

The expansion of blockchain-based platforms has accelerated a paradigm shift in digital financial systems. In this scenario, Ethereum is at the forefront due to its programmability and broad adoption. However, the blockchain user’s pseudonymity is challenging for regulatory compliance, fraud detection, and user behavior analysis. In this sense, we investigate how to classify blockchain users, based solely on their transactions’ attributes. We jointly use a computational intelligence method and machine learning classifiers to perform the profile classification of Ethereum users as professional or common accounts. Furthermore, we combine machine learning classifiers with a simulated annealing optimization algorithm to enhance classification results. Our results show that it is a promising combination, improving recall and AUC-ROC curve by at least 6% and 7%, respectively, over the previous state of the art.