Interaction Feature
A column built by combining two others — typically their product or ratio — so a model can use an effect that depends on both.
is_premium × discount_pct is an interaction. So is price / square_metres, which is the highest-yield shape on tabular data, because a raw price means very little until you divide it by what you're getting.
You build these because model families can't. A linear model adds its inputs, so it cannot represent an effect that only appears in combination. A tree cuts one column at a time along an axis, so it can approximate a ratio with a staircase of dozens of splits and never compute it exactly.
Two cautions. Generating every pairwise product explodes fast — forty columns gives 780 pairs — and selecting the best of thousands of candidates against one validation set will find winners in pure noise. And a ratio needs a guard for a zero denominator, which is more common in real data than in the design.