Underfitting
A model too simple to capture real patterns, performs poorly on both train and test.
- High error on training set, not just test set.
- Model class is not flexible enough for the data.
- Fixes: more capacity, richer features, less regularization.
When to use: Suspect when training error is also high, not just the gap to test.
Example: A straight line fit to clearly U-shaped data: train R² = 0.05, test R² = 0.03. Adding a quadratic feature lifts both above 0.7.