Identification of Suspicious Bitcoin Network Nodes by Big Data Analysis Methods

Abstract

Purpose: to analyze existing machine learning models that allow identifying suspicious addresses of the Bitcoin network, to develop a modern effective scoring model for identifying suspicious addresses. Methods: collecting and analyzing data on addresses and transactions of the Bitcoin network, identification of patterns of illicit activity of addresses, developing and experimental verification of machine learning models aimed at identifying suspicious addresses of the Bitcoin network using related transactions. Practical relevance: the analysis of common data sets and machine learning models related to the identification of suspicious Bitcoin network addresses was carried out, data on transactions related to a representative set of addresses was collected. Machine learning models have been built to identify suspicious addresses based on the collected information. Experimental approbation of the models was carried out. It is established that the best result is obtained by a model using gradient boosting. This model demonstrates more efficient operation compared to existing analogues.