Machine Learning Model for Detecting Fraudulent Transactions on the Ethereum Blockchain

Abstract

Cloud-based infrastructure can offer the required processing and storage capacity to manage massive transaction data. Cloud services increase centralization by relying on a single cloud provider, which may expose risks. Even though the cloud has a strong identity and access management system to control access, security issues might still arise. We will leverage the potential of blockchain technology along with the cloud services' scalability and adaptability to tackle these issues. Although these approaches show great potential, the issue still lies in the constant evolution of fraudulent tactics within a dynamic Ethereum ecosystem. This work combines blockchain technology with machine learning algorithms to detect anomalies in Ethereum transactions. There are various scenarios in which these scams happen, including tracking actions and monitoring transaction data. It is observed that the XGBoost algorithm outperforms with an accuracy of 99.39%. Moreover, an application for cryptocurrency transactions is integrated with the fraud detection module. As a result of the experience, cryptocurrency ecosystems already have reliable fraud detection mechanisms in place. The validation metrics exhibit a similar range, indicating that the models are not over-fit. The results show that the SMOTE oversampling techniques improve the classification F1 score levels to 98.61 with an AUC of 100%. These techniques offer a 50/50 class balance for detecting Ethereum transaction fraud.