Ethereum smart contract exploitation detection using machine learning

Abstract

Vulnerabilities in Ethereum smart contracts may be exploited by malicious actors for financial gains. While many vulnerability detection tools are available, these tools are not perfect and vulnerable smart contracts may still be deployed into the Ethereum blockchain. As such, the detection and identification of malicious transactions becomes important for contract owners and the community. In this project, we propose the use of anomaly detection machine learning algorithms to detect malicious transactions based on information recorded on the blockchain. Malicious transactions are considered anomalies and are generally uncommon as compared to benign transactions. By grouping existing smart contracts of similar functionalities, we can build a machine learning model using historical transactions information from these smart contracts and apply it to detect future malicious transactions. We will also evaluate the effectiveness of our approach on past exploitations in the Ethereum main net and present the results.