Concept - Underfitting

Definition (s.333): large training and testing errors.

In the bias-variance picture (s.345): the high bias, low variance bullseye (points tightly clustered but away from the truth) is labelled Underfitting; on the bias/variance/test-error plot (s.346) it is the left side (high bias) at low model complexity.
Example (s.334): a straight line fitted to the Happiness vs Wealth data.
Ensembles that reduce bias (see Concept - Boosting) address it.

Khaled Notes