Core Concepts
Target Column Concept
Understanding target variables and target-aware reviews
In supervised machine learning, every training dataset contains a target column — the specific column representing the outcome variable or label your model is being trained to predict.
Examples of Target Columns
- Classification:
survived(Titanic survival),churn_label(Customer churn),is_fraud(Credit card fraud). - Regression:
median_house_value(California housing),sale_price(Real estate),demand(Sales forecasting).
Why Declaring Target Column Matters
When you invoke fs.review(dataset, target_column="survived") or fs.analyze(dataset, target_column="survived"):
- Featuresmith evaluates general statistical quality across all features.
- It unlocks Intelligent Leakage Detection, comparing every feature against the declared target column to catch strong correlations, timestamp anomalies, and outcome clones.
If target_column is omitted, Featuresmith executes all statistical quality checks but skips target-specific leakage detection.