Bias

• high train and validation error
• underfitting

Variance

• high validation error
• overfitting

Bias Variance Tradeoff

Goal: Model with low bias, high variance

= Variance + Bias^2

TP, FP, F1

ROC Curves