Machine Learning Based Bitcoin Address Classification

A bitcoin address is required for trading and maintaining pseudonymity for the owner. By exploiting this pseudonymity, various illegal activities are conducted around the world. To detect and deter illegal transactions, this paper proposes a method of identifying the characteristics of bitcoin addresses related to illegal transactions. We extracted 80 features from bitcoin transactions. Using machine-learning techniques, we successfully categorized addresses involved with illegal activities with a \(\sim \)84% accuracy. We also examined the address features most affecting classification performance and compared two machine-learning models. By applying the majority voting to the classification results of bitcoin addresses associated with a particular transaction, it will be possible to determine which category the transaction belongs to.