What Is Underfitting?

Underfitting is when a machine learning model is too simple to capture the real patterns in its training data, so it performs poorly on both the data it learned from and new data it has not seen. It is the opposite of overfitting. An underfit model has not learned enough, usually because it lacks the complexity, features, or training it needs.

How underfitting happens

Underfitting shows up as a model that is weak even on its own training data. Because it never captured the underlying pattern, it makes the same kinds of errors everywhere. The usual causes are a model that is too simple for the problem, too few useful input features, or training that was stopped too early.

It is the mirror image of overfitting. An overfit model learns its training data too closely, including the noise, and fails on new data. An underfit model fails to learn the signal in the first place. Good model building aims for the middle, complex enough to capture the real pattern, without memorizing the noise around it.

How to fix underfitting

Fixing underfitting means giving the model more to work with. That can mean using a more capable model or architecture, adding features that carry more useful information, training for longer, or easing constraints that were holding the model back. The goal is to let it capture the genuine structure in the data.

Because underfitting is visible directly in poor training performance, it is often easier to spot than overfitting, where a model can look strong in training and only fail later on new data.