Automation and smart materials in detecting smart contracts vulnerabilities in Blockchain using deep learning

Abstract

Smart contracts are a vital component of applications being built by blockchain or distributed ledger technology. Smart contracts consists of computer code that composed set of rules agreed upon by the involved parties. When these predefined conditions or rules are satisfied by the transactions initiated by parties, the smart contract executes itself to complete the transaction. Normally, while performing transactions with agreements involving multiple parties, a third party have to be involve to verify all the information, which makes it complex and time consuming process. Smart contracts simplify this process by eliminating the third party and automating the process, enabling the stakeholders to perform transactions directly with each other. Smart contract itself is replicated among multiple nodes of a blockchain there by giving benefits of immutability, security and permanence. Most smart contracts are written in Solidity programming language due to its simplicity and conciseness. Attackers are attracted by the popularity of Solidity language and its vulnerability possibilities. For example, in the 2016 year, 60 million US dollars was theft by Decentralized Autonomous Organization (DAO) attack due to vulnerabilities present in smart contracts. There are few existing tools and papers on this area to find vulnerabilities on smart contracts but taking more time to predict. Thus, this paper presents different Deep learning techniques with satisfactory results to identify smart contract vulnerabilities which are Re-entrancy, Denial of Service (DOS) and Transaction Origin using binary, multi class and multi label classification techniques.