The interesting finding is usually a derived value that was not in the data: a rate, a duration, a flag, a bucket. Good features turn a question into a column you can group by, which is why the best analysis work is spent here rather than on models.
About 2 min of reading
Before this new lesson, one small question about the lesson you just finished. Recall it, do not look it up.
Build a duration feature and test it.
Duration between two dates
Before you run it, say out loud what it will print. Then open the answer.
Bucket with a stated meaning
Before you run it, say out loud what it will print. Then open the answer.
Before you write anything, put these lines back in the right order. Click a line to move it up one step. This support step is easier than writing code from scratch, and it is here to make the last example stick.
Read the first example one line at a time. A line that prints shows what appears on screen; the rest set things up quietly.
A feature computed with information from after the event being predicted is a leak, and it makes a model look excellent on history and poor on new data. Compute every feature from data available at or before the event, and check the ordering explicitly.
Twenty weak features hide one strong one and make every later step slower to reason about. Write down the three features you think carry the signal and check whether they answer the question before adding more.
'Did this customer return?' becomes a boolean column. 'How long between orders?' becomes a number. Once the question is a column, the answer is a group and a mean, which you already know how to compute.
Bucket ages or sizes only when the buckets mean something, like age bands a policy already uses. Buckets invented to make a chart look nice move the conclusion, and the same data with different edges gives a different story.
A new feature is a claim. Check it on cases you already know the answer for: a known duration, a known rate, a known flag. A feature that is wrong on a known case is wrong everywhere.
Each one is defined here on this site. Open the entry for the full definition and the other words it points at.