CTDM: cryptocurrency abnormal transaction detection method with spatio-temporal and global representation

With the rapid advances in computing and networking technologies, there have led to the creation of a novel and booming set of payment services, known as cryptocurrencies or digital tokens. Many are available for exchanges worldwide, inviting investors to trade with costs, quality, and safety that vary widely. Nevertheless, Blockchain transaction data have complex time and space dependencies, and historical transaction data reflect the transaction trends of cryptocurrencies to a certain extent, thus identifying the illegal behaviors of transactions such as money laundering more at the earliest. In this article, we propose a novel cryptocurrency abnormal transaction detection method with spatio-temporal and global representation, namely CTDM. CTDM combines EvolveGCN with MGU and global representations to achieve better performance. In addition, CTDM needs fewer learning parameters through MGU, which leads to less training time. Experimental results show that the proposed CTDM method outperforms SOTA Blockchain abnormal transaction detection methods.