Abstract: Blockchain is one of the most advanced technologies that play an important role in many different fields such as healthcare, capital markets and logistics. Among the many existing blockchain platforms, the integration of the Turing-complete virtual programming engine with the blockchain makes the Ethereum blockchain one of the most paramount infrastructures for various types of applications, including but not limited to cryptocurrency trading, smart contracts, decentralised finance and metaverse. Nevertheless, Ethereum like many other computing systems, has fallen victim to vector attacks that exploit its vulnerabilities and have catastrophic consequences. Out of the need to protect Ethereum from such attacks, this paper proposes a novel deep learning model based on convolutional neural networks. The proposed model treats the transaction, which is the atomic entity in this platform, as a stochastic time series and then develops two specific task layers that are compatible with the traditional CNN architecture. The first layer is responsible for detecting the seasonal characteristics of the transactions, while the second layer is used for detecting the trend. These two layers are integrated with the traditional architecture to form a powerful temporal CNN architecture that can classify different types of attacks. The performance of the proposed model was evaluated from a different perspective using real transactions collected from the Ethereum main-net network. The results of the comprehensive evaluations show the ability of the proposed model to perfectly identify malicious transactions in the Ethereum blockchain.
Keywords: Ethereum blockchain, sandwich attack, front-running, block stuffing, temporal CNN.
| DOI: 10.17148/IJARCCE.2022.11823