In the present study, two hybrid models are implemented, which combine machine learning algorithms with classical econometric models to forecast Ethereum volatility. The first model utilizes recurrent neural network structures in conjunction with forecasts from a GARCH(1,1) model, while the second model applies SVM for the estimation of GARCH models. It is found that both models outperform their respective base models in volatility forecasting. Additionally, the efficiency in risk measurement is evaluated using Value at Risk, in which only the second model demonstrates greater effectiveness in risk management compared to the base model.