Detecting Malicious Ethereum Entities via Application of Machine Learning Classification

Abstract

Malicious activities such as scams and frauds have imposed high costs for financial systems. The advent of blockchain-based cryptocurrencies such as Ethereum provides unprecedented characteristics. On one hand, the pseudonymity of the blockchain allows criminals to hide their actual identities, which is an appealing feature for conducting malicious activities. On the other hand, the public data of blockchain sets forth the opportunity for comprehensive forensic analysis. In this paper, we present a novel framework to identify malicious entities in the Ethereum blockchain network. The proposed framework composes of an efficient method for extracting a set of features from the Ethereum blockchain data to represent transactional behavior of entities. Our proposed solutions for detecting malicious entities employ variations of Logistic Regression, Support Vector Machine, Random Forest, and other ensemble methods such as Stacking and AdaBoost Classifier. The ensemble methods show high performance with F <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</inf> score of 0.996 in average. The results also imply that the proposed method of feature extraction is fairly efficient in presenting the network characteristics.