Overfitting happens when a machine learning model learns its training data too closely — including its noise and quirks — and as a result performs poorly on new, unseen data.
What it means in plain English
A good model learns the general patterns in its training data. An overfitted model memorises the specifics instead, including random noise that will not appear in new data. It looks brilliant on the data it was trained on but fails when faced with anything new, because it learned the examples rather than the underlying rule.
It is one of the most common problems in machine learning, and much of the craft of training good models is about preventing it.
A simple example
Imagine a student who memorises the answers to a specific practice exam rather than understanding the subject. They ace that exact test but fail the real one with different questions. An overfitted model does the same thing.
Why it matters
Recognising and preventing overfitting is central to building models that actually work in the real world. Techniques like using more data, simplifying the model, and testing on held-out data all exist largely to combat it.
Related terms
- Underfitting — the opposite problem, learning too little.
- Training Data — the data a model can overfit to.
- Model — what overfits when trained poorly.
Frequently asked questions
What is overfitting?
Overfitting is when a model learns the training data too closely — including its noise — so it performs well on training data but poorly on new, unseen data.
How is overfitting prevented?
Techniques include using more data, regularisation, dropout, simpler models, and validating on held-out data to catch it early.
