Underfitting happens when a machine learning model is too simple to capture the patterns in its data, so it performs poorly even on the training data itself.
What it means in plain English
Underfitting is the opposite of overfitting. Here the model has not learned enough — it is too simple, or was not trained long enough, to capture the real structure in the data. It misses important patterns and makes weak predictions across the board, not just on new data.
Both underfitting and overfitting are failures to generalise well; good model training aims for the balance between them.
A simple example
Trying to predict house prices using only the number of bedrooms would underfit — the model is too simple to capture everything that really drives price, like location and size, so it performs poorly for everyone.
Why it matters
Understanding underfitting, alongside overfitting, explains the central balancing act of machine learning: a model must be complex enough to learn the real patterns but not so complex that it memorises noise. Getting that balance right is much of what training a good model involves.
Related terms
- Overfitting — the opposite problem, learning too much.
- Model — what underfits when too simple.
- Feature — adding the right features can fix underfitting.
Frequently asked questions
What is underfitting?
Underfitting is when a model is too simple to capture the patterns in the data, so it performs poorly on both training and new data.
How is underfitting fixed?
By using a more capable model, adding useful features, training longer, or reducing excessive regularisation so the model can learn the underlying patterns.
