A Federated Machine Learning Method for Malicious Smart Contract Detection

Abstract

Blockchain adoption has significantly expanded in recent years, with the emergence of smart contracts facilitating practical applications in many domains. Since smart contracts can execute cryptocurrency transfers, malicious users have started implementing fraudulent smart contracts to deceive blockchain users and steal their funds. To mitigate this issue, this paper proposes an approach to detect fraudulent smart contracts leveraging Federated Machine Learning. Our approach generates an image representation for each smart contract by extracting the opcodes and assigning a unique RGB pixel. We utilized two publicly available datasets, including malicious and trustworthy smart contracts, to train multiple models on non-independent and Identically Distributed data to better represent a real-world scenario, achieving interesting results in accuracy. To the best of our knowledge, this article represents the first approach in the unique identification of fraudulent smart contracts using opcodes with a specific color, also leveraging a Federated Machine Learning approach in the blockchain environment.