부산대학교 도서관

Currency transactions in blockchain are affected by the characteristics of blockchain, and many illegal transactions need to be detected to improve security. This paper first briefly introduces currency transactions and illegal transactions in blockchain. Then, a detection method was designed to combine the Bagging algorithm and graph convolutional neural network (GCN) algorithm, i.e., Bagging-GCN. Experiments were conducted on the Elliptic dataset. The results showed that compared with traditional machine learning methods such as logistic regression and support vector machine, the Bagging-GCN method better distinguished legal and illegal transactions. In addition, regarding feature input, the effect of using all features was better than that of only using the first 94 features. In the case of sample imbalance, the Bagging- GCN method always maintained a stable detection performance. In contrast, the detection performance of the GCN method declined rapidly with the aggravation of sample imbalance. Finally, when the number of weak classifiers used was 6, the Bagging-GCN method had the best detection performance, with an accuracy of 0.937, an F1 value of 0.972, and an area under the curve of 0.956. The results prove the reliability of the Bagging-GCN method in detecting illegal transactions and its application potential in the actual blockchain.