A method includes acquiring blockchain data that includes transactions between a plurality of blockchain addresses. The method includes labeling a set of the blockchain addresses as fraudulent and generating a graph data structure based on the blockchain data. The method includes calculating a set of scoring features for each blockchain address, where each set of scoring features includes a graph-based scoring feature. Calculating the graph-based scoring feature includes calculating a number of transactions associated with the blockchain address in the graph data structure. The method includes generating a scoring model using sets of scoring features for the blockchain addresses that are labeled as fraudulent and generating a trust score for each blockchain address using the scoring features and the scoring model. The trust score indicates a likelihood that the blockchain address is involved in fraudulent activity. Additionally, the method includes sending a requested trust score to a requesting device.