Feature Engineering and Preprocessing

Feature Engineering and Preprocessing

Overview

Transform raw data into inputs a model can learn from.

Upstream context: perform basic EDA and solid Data Wrangling before heavy feature work; otherwise you may encode noise or leakage.

Techniques

• Numeric: scaling/normalization
• Categorical: one-hot encoding, embeddings
• Text: bag-of-words, TF-IDF, tokenization
• Images: resizing, normalization, augmentation

Example

• Convert timestamps into hour-of-day and day-of-week

Checklist

• Fit preprocessing on training set only
• Save preprocessing steps to apply at inference
• Document each transformation (why + reversible?)
• Avoid leaking target info (see Model Evaluation & Metrics)

Resources

• scikit-learn pipelines