Data Mining-Based Ethereum Fraud Detection

Abstract

The popularity of blockchain-based currencies, such as Bitcoin and Ethereum, has grown among enthusiasts since 2009. Relying on the anonymity provided by the blockchain, hustlers have adapted offline scams to this new ecosystem. As a result, Ponzi schemes are proliferating on Ethereum, dressed up as secure investment schemes. They reward early investors with funds from the later ones before collapsing, leaving the last investors empty handed. Illegal in the offline world, they are creating thousands of victims on Ethereum, while stealing millions of dollars worth of ether. We use data mining to provide a detection model for Ponzi schemes on Ethereum, improving over prior work. We built a dataset of likely benign Ethereum smart contracts, in addition to known Ponzi scheme smart contracts, and designed features based on their compiled code and transactions. Using Weka to benchmark several classification algorithms, we obtained models that achieve both high precision and high recall. Our 0-day model can be used as soon as a smart contract is uploaded on the blockchain. The full-feature model continued to show high performance for almost 250 days. A detailed analysis on top-strength features provides novel perspectives on Ponzi scheme behavior.