Detection of Fraudulent Entities in Ethereum Cryptocurrency: A Boosting-based Machine Learning Approach

Due to the rise in the use of crypto-currencies, such as Bitcoin and Ethereum, the fraud activities in the financial sector are increasing at the same pace. It becomes very challenging to detect frauds in crypto-currencies, because of it's distributed and anonymized nature and not having central control. In this paper, we exploited the decision-tree based machine learning model using a boosting approach, particularly XGBoost, to identify the fraudulent addresses in the Ethereum crypto-currency. To select a best decision tree and learning approach on our fraud dataset, initially, we chose four highly performed decision tree learning approaches including CART, random forest, LGBM, XGBoost, and applied a cross validation mechanism to select the top one based on accuracy. Among CART, random forest, gradient boosting tree, we selected the XGBoost model as the final model and tuned it for the best hyper parameters. Finally, we built the model on 80% of the training data, which has produced accuracy of more than 96% on test data. Further, the model is highly efficient to work in a real environment, as proved by running extensive experiments.