accuracy

Is a common metric in classification.

Fails when classes are high imbalanced. For these cases f1-score is more appropriate

$$A=\frac{1}{n}\sum _{i=1}^{n}1({\hat{y}}_i=y_i),$$

where

• $n$ is the number of observations;
• $1$ is the indicator function
• $\hat{y}$ is the predicted value
• $y$ is the observed value