Research on Dynamic Detection of Vulnerabilities in Smart Contracts Based on Machine Learning

Abstract

The proliferation of smart contracts has led to a surge in hacking attacks, resulting in substantial financial losses and undermining the healthy growth of the blockchain ecosystem. To mitigate these challenges, this paper introduces a dynamic vulnerability detection approach for smart contracts leveraging machine learning techniques. The proposed method involves the extraction of opcode sequence features through a combination of the N-gram model and a weight penalty mechanism. The core objective is to identify vulnerabilities within deployed smart contracts by analyzing the opcode sequences during their dynamic execution. This approach falls under the category of dynamic detection, aiming to ensure the integrity and security of blockchain-based systems.