A model that predicts the average price for every house, regardless of size, location, or age, is underfitting. It has not learned enough structure to be useful. Underfitting happens when the model is too simple, the training time too short, or the features too weak to capture the pattern in the data.

The symptom is easy to spot: poor performance on the training set itself. Unlike overfitting, where a model memorizes noise and fails on new data, underfitting fails everywhere. A straight line trying to fit a curve will miss both the training points and the test points.

Typical causes

• Model capacity too low for the complexity of the problem
• Too few training epochs or a learning rate that stalls
• Missing or poorly engineered input features
• Excessive regularization that constrains the weights too much

The fix is usually to add capacity, train longer, or feed the model better inputs. Sometimes the problem is not the model at all but the data: if the signal is buried in noise, no architecture will rescue it. Diagnosing underfitting early saves weeks of wasted tuning.