What is Feature Engineering?
Feature Engineering — The process of using domain knowledge to extract features from raw data for use in machine learning.
Feature engineering transforms raw data into meaningful inputs for ML models. Examples include extracting day-of-week from timestamps, calculating customer lifetime value from transaction history, or computing moving averages from time series. Good features often matter more than model choice.
Frequently Asked Questions
Is feature engineering still relevant with deep learning?
Less so for unstructured data (text, images) where deep learning learns features automatically. For structured/tabular data, feature engineering still significantly impacts model performance.
What makes a good feature?
Good features are predictive of the target, available at prediction time, not leaky (don’t contain future information), and interpretable. Domain expertise is critical for identifying them.
Can feature engineering be automated?
Partially. Tools like Featuretools and AutoML platforms can generate candidate features automatically. But domain-specific features crafted by experts typically outperform automated approaches.