Thesis: 'Classification of Bitcoin transactions based on supervised machine learning and transaction network metrics.' - Distinction
Thesis Abstract: "The Bitcoin currency is a publicly available, transparent, large scale network in which every single transaction can be analysed. Multiple tools are used to extract binary information, pre-process data and train machine learning models from the decentralised blockchain. As Bitcoin popularity increases both with consumers and businesses alike, this paper looks at the threat to privacy faced by users through commercial adoption by deriving user attributes, transaction properties and inherent idioms of the network. We define the Bitcoin network protocol, describe heuristics for clustering, mine the web for publicly available user information and finally train supervised learning models. We show that two machine learning algorithms perform successfully in clustering the Bitcoin transactions based on only graphical metrics measured from the transaction network. The Logistic Regression algorithm achieves an F1 score of 0.731 and the Support Vector Machines achieves an F1 score of 0.727. This work demonstrates the value of machine learning and network analysis for business intelligence; on the other hand it also reveals the potential threats to user privacy. "