🤖 ML Models
ML - Feature Engineering for Crypto
The process of transforming raw price/volume data into meaningful inputs for ML models. In finance, features are the "signals" the model uses to make predictions. Feature engineering is widely considered more important than model choice.
2
Minutes
3
Concepts
+15+30
Read+Quiz
1
Signal Forge's 27 Features (current)
Returns (4 features)
return_1h, return_4h, return_24h, return_7d
What: Price change over different lookback windows
Why: Captures momentum at multiple timescales
Volatility (2 features)
volatility_24h, volatility_7d
What: Rolling standard deviation of returns
Why: High volatility = uncertain, low volatility = calm (often precedes big moves)
Volume (2 features)
volume_change_24h, volume_sma_ratio
What: Volume relative to recent history
Why: Volume confirms price moves — big move + big volume = real, big move + low volume = suspect
Technical Indicators (5 features)
rsi_14          Relative Strength Index (overbought/oversold, 0-100)
macd            Moving Average Convergence Divergence (trend strength)
macd_signal     MACD signal line (crossover = trend change)
bb_upper        Bollinger Band upper (price at 2 std devs above mean)
bb_lower        Bollinger Band lower (price at 2 std devs below mean)
Crypto-Specific (3 features)
funding_rate_proxy    Estimated from OHLC spread (basis between spot and implied future)
volatility_regime     Categorical: low/normal/high/crisis
btc_correlation       Rolling 30-day correlation with BTC (altcoins only)
Time Features (2 features)
hour_sin, hour_cos    Cyclical encoding of hour-of-day
Why: Crypto has intraday patterns (Asian/European/US session effects)