Fraud Detection In Ethereum Transactions: A Machine Learning Approach

Abstract

This study explored the use of machine learning models to detect fraudulent transactions made over Ethereum, a type of cryptocurrency. The dataset included data from 9840 transactions which included both fraudulent and legitimate transactions and included various parameters such as average, minimum and maximum amounts received and sent, the number of unique addresses to and from which ether was sent and received, the time duration between the transactions and the remaining balance after the transactions. The transaction amount lies in a range of 12000.0 to 52000.0. Four machine learning models, namely Decision Tree, Random Forest Classifier, Catboost Classifier and XGBoost Classifier, are evaluated for fraud transaction detection. XGBoost has emerged as the most promising model, achieving near perfect accuracy (99.65%) and high precision, recall and F1-scores. Notably the robust performance of XGBoost suggests it’s potential for accurately detecting fraudulent transactions in Ethereum. However, comparatively weaker performance is observed in models like the Decision Tree and Random Forest. This study underscores the significance of machine learning in enhancing fraud detection in Ethereum transactions and advancement of the field of cybersecurity in blockchain technologies by providing a useful and scalable fraud-detecting instrument.