This paper presents the study of fraud detection on the Ethereum blockchain system using several machine learning techniques for predictive analysis of fluctuations in the stable coin markets, e.g., USDT, USDC, and DAI. The system recommended as “Crypto-Macer” comes with the AdaBoost algorithm for criminal transaction detection, whereas XG Boost is employed in forecasting the price of stablecoins. When looking at the issue of fraud on the Ethereum dataset, a number of variables are accounted for such as the number of transactions, amount of money transacted, time taken between two transactions and the number of transactions done per user within a given period. In predicting the price of stablecoins, models describing past prices and trading volumes are used. In order to achieve better performance of her model in practice, which was a concern, feature engineering and model tuning were carried out. Within this research it was shown, that AdaBoost model accuracy related to fraud detection and XGboost model accuracy related to stablecoin price predicting tend to be close to one another—it is possible to make positive advances in economically wasteful blockchain-based systems using machine learning.