Random Forest Signals for Intraday Bitcoin Trading: Features and Caveats

Summary

This project applies a Random Forest classifier to intraday BTC/USD data to produce directional signals from technical features. It uses two years of one-minute OHLC observations and inputs including returns, percentage changes, RSI, ADX, moving-average relationships, correlation, and return volatility. The predicted class represents the direction of the next return. The author describes choosing the Gini criterion, unrestricted tree depth, and 11 estimators after comparing estimator counts.

The article reports a Sharpe ratio of 4.47 and total return of 367.05%, and says the strategy outperformed buy and hold. These are backtest results as reported by the author; the supplied text does not show the full performance table, and it provides limited information about validation design, trading costs, or out-of-sample robustness. Volume is excluded because the data source returned zero volume for many records. The author also warns that frequent trading makes the model impractical in its current form and frames it as a research starting point rather than a ready-to-use strategy.

Key ideas

• A Random Forest classifier maps technical indicators from one-minute BTC/USD bars to the next return’s direction.
• The feature set includes momentum, trend, moving-average relationships, and volatility measures.
• The author reports strong backtest metrics versus buy and hold, but detailed validation evidence is absent from the text.
• Unreliable volume data led to excluding volume and volume-derived indicators.
• The author cautions that the high trading frequency makes the strategy impractical without further development.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.