A Feature Engineering-Driven Approach to Detecting Fraud in Ethereum Transactions with Ensemble Models

The Ethereum blockchain, known for its decentralized and open-source nature, has revolutionized online transactions. However, vulnerabilities within Ethereum's architecture have resulted in several security breaches. By analyzing transaction data, malicious activities can be detected, helping to mitigate cyber threats such as phishing, Ponzi schemes, and eclipse, Sybil, and DDoS attacks. Machine learning-based anomaly detection has proven effective across various fields, and this paper introduces a fine-tuned ensemble machine learning model to detect fraudulent transactions on the Ethereum platform. To optimize the identification of fraudulent transactions, XGBoost and Random Forest algorithms are utilized to extract critical features from the transaction dataset. Additionally, data resampling techniques are applied to counteract overfitting. The proposed framework operates in two phases: the first phase assesses the effectiveness of different machine learning models, while the second phase develops an ensemble model based on these findings. The framework's performance is benchmarked against the baseline models and state-of-the-art methods, demonstrating superior results. Testing reveals that the ensemble model achieves an accuracy of 99.4% and a Matthews’s correlation coefficient of 94.9%.