Detecting Vulnerable Smart Contracts by Security Risk Estimation

Abstract

Detecting vulnerable smart contracts has a direct effect on blockchain security because it helps users avoid using these contracts. In this study, the problem of vulnerability risk for blockchain smart contracts is introduced. Moreover, an effective criterion for its estimation is devised. With this criterion, to estimate the risk of an unknown smart contract, linear discriminant analysis of smart contracts and distances to their nearest neighbors were exploited. Although deep learning is not used in the proposed criterion and it requires little training data, it provides a realistic risk estimation of smart contracts. The experiments conducted on a real-world dataset of Ethereum blockchain smart contracts, including both vulnerable and safe contracts, show the acceptable performance of the proposed criterion. Moreover, the performance of the proposed criterion is superior to that of existing criteria in other areas of risk estimation.