Leveraging Deep Learning for Enhancing Blockchain Smart Contract Security: A Systematic Literature Review

Although blockchain smart contracts hold great promise for decentralised applications, the security threats against them are quite prevalent and can be very expensive. Deep learning techniques have started to show prominent prospects in their detection. Despite the increased concern over these weaknesses, the current studies have mainly focused on detection methods based on machine learning without considering the role of deep learning techniques in smart contract vulnerability detection. This paper presents a systematic literature review of 16 recent studies strictly related to deep learning-based smart contract vulnerability detection. This study analysed the various deep learning models applied, covering variants of the LSTM, CNN, GNN, and hybrid models. Their performance is assessed with performance metrics, and their targeted vulnerabilities are considered; their limitations and weaknesses are addressed. The study reveals a promising trend in this review, where custom datasets and models tailored to specific vulnerabilities often yield high accuracy rates. However, there are challenges to generalising these approaches across diverse vulnerabilities and datasets. This review gives the research community an overview of the current landscape, identifies the methods used, and identifies the types of vulnerabilities trending in the current approaches to detection.