Because of its convenience,anonymity,globality and high mobility,Bitcoin provides an ideal tool for criminals to use it as amedium of value transmission to engage in criminal activities,resulting in a large number of abnormal transactions such as extortion,money laundering,illegal drugs and weapons trading. The traditional method?of anomaly address recognition based on supervision cannotfully and accurately reflect the relationship between addresses due to the single transaction information, so the recognition rate of anomalyaddress is low.?
Therefore,we propose a Bitcoin anomaly address recognition method based on transaction network feature enhancement.This method converts Bitcoin transaction data into a complex network and extracts network features,and proposes a node importancefeature construction method based on improved PageRank. According to Bitcoin transaction features,the Bitcoin transaction quota andfrequency correlation are introduced to obtain new PR values and add them to the feature collection. By comparing different machinelearning methods,we can get the best prediction model and improve the classification effect of the detection model. Compared with thetraditional detection methods, the model combined with network information has a better detection performance. Among them, theXGBoost classifier has the best performance. The F1 score increases from 0. 83 to 0. 94,and the AUC value increases from 0. 88 to 0. 95.