05·Feature engineering

Introduction

Feature engineering is the deliberate construction of new variables that expose data structure to the model. The lesson covers classical techniques (polynomial features, interactions, ratios, bucketing), domain-specific ones (date, geography, text, time-series — lag and rolling), per-group aggregations, frequency/target encoding with K-Fold discipline, PCA as a dimensionality reducer, automated generation (Featuretools/DFS), the curse of dimensionality, and reproducibility via the sklearn Pipeline. Guiding rule: feature engineering MUST be part of the pipeline, fit on train only, to prevent leakage.