Ethereum Smart Contracts Vulnerabilities Detection Leveraging Fine-Tuning DistilBERT

Abstract

In recent years, the proliferation of blockchain technology has led to the widespread adoption of smart contracts, particularly on the Ethereum platform. However, these smart contracts are prone to vulnerabilities that can be exploited, leading to significant financial losses and security breaches. This paper explores the application of Distil-BERT, a distilled version of the BERT (Bidirectional Encoder Representations from Transformers) model, to detect vulnerabilities in Ethereum smart contracts. By fine-tuning Distil-BERT on the SB-curate dataset, a curated dataset of vulnerable Solidity smart contracts, we achieve a performance accuracy exceeding 97%. We provide natural language processing (NLP) techniques that can be leveraged to enhance the accuracy and efficiency of these methods. Our experiments show that Distil-BERT can effectively identify common vulnerabilities, such as reentrancy and integer overflows, outperforming traditional static analysis tools. This research contributes to developing more secure smart contract systems and offers insights into integrating advanced machine learning models in blockchain security.