Overfitting
Overfitting is a condition in which a model is fitted too closely to the data it was trained on, so it performs worse when making predictions on new data.
On this page
Source content checked . English translation published 2026-10-27.
In business automation
A model that performs very well on training examples may fail as soon as an inquiry arrives with different wording. Overfitting is this condition: excessive fit to the training data that harms predictions on new data. In the guide, the term appears alongside machine learning and fine-tuning.
Examples
These examples illustrate a possible workflow. Adapt the connections and rules to your own systems.
Near-perfect classification on an old dataset, poor results on a new week
An inquiry classification model succeeds almost every time on the test dataset. When inquiries from a new week arrive, accuracy drops. The team checks whether the model memorized old wording instead of learning the category.
Relevant tools
- Replicate
- When running a ready-made model, overfitting may have already occurred during its training. Test it on your own examples. This does not rate any particular model's quality.
What to check
- Choosing a model based only on its score on the training dataset hides overfitting.