Lesson 2.4 · 23 min · video lesson
Feature Engineering: Turning Raw Data into Signal
Why can the same simple model go from an average error of 40.8 to 6.7 without changing a single line of the algorithm, just by changing how we describe the data?
In short: Feature engineering is turning raw data into input columns (features) that make the pattern easy for a model to learn: filling missing values, encoding categories, scaling numbers, transforming skewed values, and creating new features from domain knowledge. Good features often matter more than the choice of algorithm, especially for tabular data. Deep learning learns many features automatically, but feature engineering still matters in production AI systems.
In this lesson
- Why the way we describe data matters
- Where feature engineering fits
- The core techniques, one by one
- Code: same model, better features
- Hand-made features vs learned features
- Scaling in detail: standardisation vs min-max
- Common mistakes and when to stop
- Worked example, step by step
- Practice: try it yourself
Key terms in this lesson
- Feature
- A measurable input variable given to a model.
- Feature engineering
- Creating, transforming and selecting features so a model can learn the pattern more easily.
- One-hot encoding
- Representing a category as a set of 0/1 columns with a single 1 marking the category.
- Standardisation
- Rescaling a feature to mean 0 and standard deviation 1 with z = (x − mean) / std.
- Imputation
- Filling in missing values with a substitute such as the median or a special category.
- Data leakage
- When training uses information that would not be available at prediction time, giving falsely good results.
This lesson is free. Sign in or create a free account to open it with its video and quiz.
← 2.3 Predicting Numbers vs Categories: Regression Compared · 2.5 Precision and Recall: Picking the Right Metric →