<![CDATA[The cryptocurrency market is characterized by fluctuating, non-stationary prices and rapidly changing volatility, which makes market-condition analysis complex. This study applies a hybrid machine learning approach combining ExtraTrees Classifier and LightGBM Regressor to support Ethereum trading analysis. Daily ETH/USDT data were collected from the Binance Public API for 1 January 2021 to 31 December 2025, totaling 1,826 raw observations. After preprocessing, feature engineering, and target formation, 1,622 observations were used and split chronologically into 70% training and 30% testing data. ExtraTrees classifies the next-period market condition into Low Volatility and High Volatility, while LightGBM estimates the next-period price change as a price delta. ExtraTrees achieved 89.12% test accuracy, exceeding the 65.91% majority baseline. On the training data, LightGBM obtained MAE of US$47.41, RMSE of US$75.68, R² of 0.9930, and Directional Accuracy of 93.04%. The outputs of both models were integrated using fusion logic with 75% volatility score and 25% magnitude score, producing Prepare to Trade, No Trade Zone (Sideways), and High Vol (No Trade) recommendations. The three-state accuracy reached 73.31%, while binary volatility accuracy reached 90.55%. SHAP was used to explain feature contributions, and the final model was implemented in a Streamlit-based Decision Support System.]]>
Copyrights © 2026
