| Abstract [eng] |
Summary Nowadays interest in bitcoin and the blockchain is very big. In 2008 when bitcoin was mentioned for the first time and bitcoin’s white paper was published, not many believed in the future of bitcoin as currency. The first years of bitcoin were not so impressive, but step by step bitcoin became more and more popular. Nowadays bitcoin gets much attention from journalists, as well as from data scientists. While journalists are publishing articles, interviews with famous people related to bitcoin, data scientists are analyzing bitcoin blockchain. Special attention is dedicated to detecting money laundering cases in bitcoin blockchain. There are many mathematical models to analyze bitcoin blockchain, transactions and to explain what kind of economic activity is represented in those. Money laundering cases are usually detected by mathematical models. There are usually used machine learning models which aim to classify, analyze transactions of bitcoin, identify suspicious transactions as well as divide it by possible economic activity. Data about bitcoin’s transactions are publicly available on the internet. The data about bitcoin’s transactions is used in this analysis. The transaction’s data is used to create graphs. The final dataset includes properties about graphs as well as information about transactions. The final dataset is clustered using k-means method, agglomeration method and density-based clustering method. Results of these methods are compared together, and the best method is chosen. Also, during the analysis, there is created and used transaction evaluation coefficient, which helps to evaluate each transactional graph separately, without clustering data first. After clustering analysis and evaluation of transactions using transaction evaluation coefficient, transactions were divided into four different groups. Each group represents common properties. Using the results of clustering, it is possible to explain which cluster is representing which economic activity. In the result, there is one cluster which has the most suspicious transactions of the block. It shows that clustering and transaction evaluation coefficient can help to predict which transaction is suspicious and should be investigated on a higher level. Also, during this analysis, it was noticed, that transaction evaluation coefficient can help to evaluate transactions separate, without processing clustering analysis first. |
