What Is Overfitting?

1 min

Overfitting is when a model memorizes the training questions instead of actually learning — it then fails on anything new.

Imagine it like this: A student who memorizes the exact answers of last year's mock exam will pass that same exam again — but will fail the real exam because the questions are worded differently. They didn't learn, they just memorized.

Examples:

  • A model that "learns" a set of photos so precisely that it fails on any new photo
  • A model that perfectly predicts past stock prices, but is useless for the future

So? A model with a perfect score on training data but a bad score on new data hasn't really learned — it just memorized. That's a big red flag!