data-feature

What problem does it solve?

Converts domain hypotheses and raw project data into a validated set of predictive features while preventing common pitfalls such as target leakage and multicollinearity, and producing an auditable feature engineering record.

Core Features & Use Cases

• Feature ideation from hypotheses: map analysis_context.md hypotheses to concrete candidate features and data sources.
• Transformations & encodings: guidance for numeric transforms, binning, one-hot/target encoding, scaling, and time-series lags/rolling features.
• Safety checks & validation: explicit leakage prevention, train/test split handling, VIF checks for multicollinearity, and recommended importance estimation (SHAP/permutation).
• Operational output: produces a documented execution log and a saved report for downstream modeling (data/docs/05_feature_engineering.md).

Quick Start

Use the data-feature skill to generate candidate features from analysis_context.md and project data, perform leakage and VIF checks, and save the feature engineering report to data/docs/05_feature_engineering.md.