Інтелектуальні методи виявлення вразливостей у смарт-контрактах Ethereum: сучасний стан і перспективи

Abstract

This paper reviews recent machine learning methods for detecting vulnerabilities in smart contracts deployed on the Ethereum blockchain. It categorizes existing approaches into seven major ML model types: CNN, RNN, GNN, and large language models. The review emphasizes the importance of feature extraction, dataset quality, and hybrid techniques that combine static and dynamic analysis. Results indicate that ML, primarily ensemble and deep learning models, significantly improves detection accuracy. The paper also outlines current challenges and trends, including explainability, real-time vulnerability assessment, and the integration of AI tools in blockchain security pipelines.