Anomaly Detection Method for Small-Scale Blockchain Transaction

Abstract

Blockchain is a distributed ledger technology that enables tamper-resistant money transfers without a trusted third party. A major application of blockchain is a cryptocurrency used for international transactions and interpersonal transactions as a low-cost and high-speed remittance method. On the other hand, blockchain has the problem that the damage of illegal transactions is more likely to be large, because the illegal transactions cannot be modified or deleted after the transactions are approved. The illegal transaction have to be detected and modified before the transaction approval to prevent the damage of it. Prior works propose the methods to detect illegal transactions through anomaly detection, because illegal transactions have different characteristics from normal transactions. Most of the prior works only focus on large-scale anomaly transactions and do not consider small-scale anomaly transactions. The accuracy of these methods for small-scale transactions is not high because there are cases where the difference in features between small-scale anomaly transactions and normal transactions is smaller than the difference between small-scale and large-scale normal transactions. However, most individual users make small-scale transactions, so preventing small-scale illegal transactions is important for system reliability and dissemination. Therefore, in this paper, we propose an anomaly detection method to detect small-scale anomaly transactions by making a subgraph that excludes the hub users who issue large-scale transaction from the user graph. We evaluate the proposed method from two perspectives, that are execution time and accuracy of anomaly detection. As a result, the execution time is shorter than the transaction approval interval, and the accuracy for small-scale anomaly transaction is improved from the existing method without decreasing the accuracy of large-scale anomaly transaction.