Feature Engineering & Signal Generation

Feature engineering transforms raw data into meaningful model inputs. NEI Predict uses automated feature generation combined with domain expertise to build signal libraries for each prediction target. Technical features for financial models: moving averages (5/10/20/50/200 day), RSI, MACD crossovers, Bollinger Band width, VWAP, order flow imbalance. Each tested for predictive power using mutual information scores. Alternative data features: social media sentiment, news event classification, on-chain metrics (crypto), satellite imagery (commodities), weather data (energy/agriculture). These often lead price by hours or days. Feature selection: from hundreds of candidates, pick top 20-40 with highest predictive value and lowest correlation to each other. Redundant features add noise. L1/L2 regularization penalizes models relying on too many weak features.