Prediction of Ethereum Blockchain ERC-20 Token Standard Smart Contract Vulnerabilities Using Source Code Metrics

Abstract

In this study, firstly, a dataset of 10,476 annotated vulnerable ERC-20 standard token smart contracts (belonging to a set of 33 common smart contract vulnerabilities) has been collected from a publicly available repository. Secondly, using the SolMet smart contract metrics measurement tool, the object-oriented software attributes (i.e., metrics) from each smart contract's source code has been extracted. Lastly, using the source code metrics and the vulnerability annotations (i.e., labels) as the input in supervised machine learning (classification) algorithms, the accuracy of each individual algorithm is evaluated against the accuracy of an ensemble classifier (namely voting). The model accuracies demonstrate the feasibility of identifying and prioritising smart contracts for further inspection prior to deployment to the blockchain network. The ensemble classifier performed better (accuracy = 0.79) compared to each classifier when used individually.