Feature Engineering
Turn raw data into useful ML features, with reliable splits, transformations, and model comparisons.
Background reading
Introduction to Feature Engineering
A model cannot use a raw column until we decide what the column means and how to represent it. A city may become indicator columns, a missing income may need a fill value, and a timestamp may become hour or day-of-week.
Those choices must be learned from training data only. If we use the test set to choose a median, category list, or feature, the model has already learned something about the data meant to test it.
This course follows one safe order: define the prediction, split the rows, inspect the training set, fit transformations on that training set, compare complete pipelines, and evaluate the chosen pipeline once on held-out data.
The focused examples run with Python, NumPy, pandas, and scikit-learn. The final workflow combines the same steps in one pipeline.
Loading course…