LGBM: a machine learning approach for Ethereum fraud detection

Ethereum is a software platform that uses the concept of blockchain and decentralizes data by distributing copies of smart contracts to thousands of individuals worldwide. Ethereum, as a currency, is utilized to exchange value worldwide in the absence of a third party to monitor or intervene. However, as online commerce grows, a slew of fraudulent activities, such as money laundering, bribery, and phishing, emerge as the primary threat to trade security. This paper proposes Light Gradient Boosting Machine (LGBM) approach for accurately detecting fraudulent transactions. It also examines different models such as Random Forest (RF), Multi-Layer Perceptron (MLP), etc., based on machine learning and soft computing algorithm for classifying Ethereum fraud detection dataset with limited attributes and compares their metrics with the LGBM approach. A comparative study of scores of bagging models is presented to know the applicability of the proposed approach. The light gradient boosting machine (LGBM) algorithms and Extreme Gradient Boosting (XGBoost) demonstrate the highest accuracies, while LGBM shows slightly better performance with 98.60% for the stated dataset scenarios. Further optimizing the LGBM with hyper-parameter tuning, an accuracy of 99.03% is achieved.