K-Means Clustering for Identifying Suspicious Patterns in Ethereum Networks

Abstract

The use of Ethereum cryptocurrency has been increasing recently, and thus there has been a growing demand for detecting fraudulent activities to protect digital financial systems. This paper seeks to apply the K-Means clustering algorithm in detecting fraud among the Ethereum transactions based on account transactions and behavior characteristics. The Ethereum Fraud Detection Dataset from Kaggle will be used to analyze fraudulent and genuine transactions. The dataset includes account addresses, number of transactions per account, transaction intervals, amount sent/ received from the account in terms of Ether, account balances, distinct sender/receiver addresses, contract calls and ERC20 Token transactions characteristics. The data pre-processing, feature selection, feature scaling and 80:20 train-test splitting will be part of the detection process. The results obtained from experimentation show that the model has an accuracy of 98.07% and F1-score of 98.08%. This confirms the ability of the method to differentiate between fraudulent transactions and genuine activities. The method allows for systematic analysis of cryptocurrency transactions and the detection of abnormal behaviors. Further verification through experimentation is needed especially due to the skewed classes in the dataset.