feature-engineeringlisted

# Feature Engineering ## References See [references/rolling-stats-reference.md](references/rolling-stats-reference.md) for the complete `RollingFeatures` constructor, all 9 available statistics, feature name generation formula, window behavior, and `kwargs_stats` usage. ## When to Use This Skill - Forecast accuracy has plateaued and you suspect better features would help - User asks about "exogenous variables", "external regressors", or "feature creation" - Time series has calendar patterns (hourly, weekly, seasonal) not yet captured - Raw datetime index is used directly instead of engineered features - User mentions feature_engine, RollingFeatures, or skforecast preprocessing - Energy/transport/outdoor domain where sunlight hours may be predictive ### When NOT to Use - **Tabular ML (non-time-series)**: Use a general feature engineering skill instead - **Deep learning forecasters** (RNNs, Transformers): These learn features internally; manual engineering adds less value - **Feature selection/importance**: This skill covers creation, not selection — use model-based selection after creating features - **Data cleaning/imputation**: Handle missing values and outliers before feature engineering ## Overview | Tool | Package | Purpose | |------|---------|---------| | `DatetimeFeatures` | feature_engine | Extract calendar features from datetime index | | `CyclicalFeatures` | feature_engine | Encode cyclical features with sin/cos | | `RollingFeatures` | skforecast | Rolling wi