Detection of vulnerabilities in blockchain smart contracts using deep learning

Abstract

Blockchain helps to give a sense of security as there is only one history of transactions visible to all the involved parties. Smart contracts enable users to manage significant asset amounts of finances on the blockchain without the involvement of any intermediaries. The conditions and checks that have been written in smart contract and executed to the application cannot be changed again. However, these unique features pose some other risks to the smart contract. Smart contracts have several flaws in its programmable language and methods of execution, despite being a developing technology. To build smart contracts and implement numerous complicated business logics, high-level languages are used by the developers to code smart contracts. Thus, blockchain smart contract is the most important element of any decentralized application, posing the risk for it to be attacked. So, the presence of vulnerabilities are to be taken care of on a priority basis. It is important for detection of vulnerabilities in a smart contract and only then implement and connect it with applications to ensure security of funds. The motive of the paper is to discuss how deep learning may be utilized to deliver bug-free secure smart contracts. Objective of the paper is to detect three kinds of vulnerabilities- reentrancy, timestamp and infinite loop. A deep learning model has been created for detection of smart contract vulnerabilities using graph neural networks. The performance of this model has been compared to the present automated tools and other independent methods. It has been shown that this model has greater accuracy than other methods while comparing the prediction of smart contract vulnerabilities in existing models.