Multi-Timeframe Forecasting of Ethereum Prices: A Comparative Study of Statistical and Deep Learning Models

Abstract

Prices of cryptocurrencies are tough to forecast due to their high volatility and susceptibility to abrupt market changes. This paper compares four models—ARIMA, Prophet, LSTM, and XGBoost—to predict Ethereum (ETH) prices on three horizons: 15 minutes, 1 hour, and 1 day. We compared all four models concerning Root Mean Squared Error (RMSE) from the historical ETH data. The outcome shows XGBoost performs best on short-term forecasting with an RMSE of 352 in 15-minute and 357 in 1-hour data, surpassing LSTM and ARIMA. For the daily prediction, Prophet shows competitive performance with an RMSE of 941, whereas ARIMA is generally stable. The findings conclude that the ideal model depends on the forecasting horizon, and for short-term trading, using XGBoost is advisable, while Prophet is advisable for longterm forecasting. The study provides valuable recommendations to investors and researchers seeking effective cryptocurrency prediction software.