<![CDATA[The decentralized finance (DeFi) and blockchain environment encounters substantial security threats, particularly complex and expensive fraudulent activities. Conventional detection methods frequently prove insufficient when dealing with enormous transaction volumes and datasets characterized by unbalanced class distributions. This research seeks to examine and evaluate the effectiveness of three widely used machine learning techniques Logistic Regression, Random Forest, and XGBoost in identifying fraudulent activities within blockchain transactions. The investigation utilized an Ethereum transaction dataset sourced from Kaggle, where the imbalanced data distribution was addressed through the application of SMOTE methodology. Performance assessment was carried out using precision, recall, F1-score, and ROC-AUC measurements on testing data. The findings demonstrate XGBoost's superiority among the algorithms, delivering an accuracy rate of 99.46%, precision of 99.69%, recall of 97.86%, and ROC-AUC score of 99.97%, while maintaining minimal false positive occurrences (only 1 instance). These results exceeded those achieved by both Random Forest and Logistic Regression models, demonstrating that gradient boosting methodologies excel at detecting intricate fraudulent behaviors. The study's outcomes offer significant contributions toward creating resilient and autonomous fraud detection frameworks. Keywords: Blockchain, Fraud, Machine Learning, Logistic Regression, Random Forest, XGBoost.]]>
Copyrights © 2026
