3.8 Feature engineering
You can turn raw columns into features that lift a model.
Before:01. Mathematics for Machine Learning02. Python — Basics to AdvancedUnlocks:05. Classical Machine Learning
Features are where domain knowledge enters the model, and good ones routinely beat better algorithms — encodings, scalings, date and text decompositions are the standard toolkit. It sits late in the data module because features compose everything learned so far. The trap with the highest cost is target encoding without cross-fold care: the label leaks into the features, validation scores soar, and production quietly returns to earth.
Work through these
Encoding: one-hot, ordinal, target, hashing
Turning categories into numbers, with four approaches that differ in how many columns they create and how much they risk leaking the target. Choosing wrongly either bloats the data or inflates the score.
Scaling and transformations, and when they matter
Putting features on comparable scales, and reshaping skewed ones. Some models care enormously about this and others not at all, and knowing which is which saves wasted effort.
Date, text and geospatial features
Dates, free text and locations each carry far more information than their raw form suggests, and extracting it is where most real gains come from. This item is the highest-value part of the topic.
Interaction terms and binning
Combining features into products, and cutting continuous values into bands. Both encode knowledge about the problem that the raw columns do not carry.
Sign in to keep your progress.
Free resources
We haven't checked most of these for screen reader use yet.
Links last checked 29 Aug 2026.
Stuck here?
Ask a mentor. A real person answers, and they can see exactly which topic you're on. Usually within a couple of working days.
Checking your session…
Topics shown in module order.