Security on Ethereum: Ponzi Scheme Detection in Smart Contract

Ethereum has many transaction security issues such as Ponzi schemes, which are hidden in a large number of smart contracts. And they are difficult to be detected. Therefore, we propose a novel multi-granularity multi-scale convolutional neural network model (MM-CNN) to detect Ponzi schemes in smart contracts. A multi-granularity method is used to compress the smart contract opcodes with similar function to obtain multi-granularity frequency data of opcodes in MM-CNN. Then, we use a multi-scale convolution kernel to extract features of frequency data. The experiments show that the frequency features are the best measurements to represent the attributes of the Ponzi scheme. In the multi-granularity method, fine-grained opcode has a stronger ability to express Ponzi attributes. The recall rate of MM-CNN on the verification set is 98.07%, which shows the effectiveness of the scheme.